{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"---\n---","metadata":{}},{"cell_type":"markdown","source":"<H1 style=\"text-align:center;\"> Actionable Insights in Cyber-physical Systems </H1>\n<H2 style=\"text-align:center;\"> Author: Andres Villalva </H2>\n<H3 style=\"text-align:center;\"> Category: Other </H3>\n\n\n# Introduction\n\nIn technology, few mistakes are as glaring as poor action selection by a robotic system. Drawing from our understanding of human physiology, we know that decision-making necessitates two key elements. The first is a perception that triggers the action, and the second is that the response is timely and proportionate to the insight. \n\nConsider the process of walking through a forest. At any given moment, a vast amount of sensory information - sights, sounds, and other environmental factors - is distilled into a single insight that guides our actions. These insights might include: \"The path is safe, continue walking,\" \"The ground looks unstable, step carefully,\" \"The surface is slippery, take smaller steps,\" or \"Something feels off, stay alert.\"\n\nThis innate ability to condense a large, complex state space into a single practical insight is a highly efficient decision-making mechanism. However, its effectiveness depends on one critical condition: that the computation process is completed within an appropriate timeframe. In robotics, as in the physical world. 'The right move, too late' is still a failure.\n\nTaking human ability as a benchmark for the speed necessary for successful interactions with the environment, a group of neuroscientists from MIT measured that the human brain can process complete images perceived by the eye in as little as 13 milliseconds [20]. In a concurrent cognitive process, the detection of auditory stimuli occurs in approximately 100-150 milliseconds [21].\n\nIt seems intuitive to suggest that cyber-physical systems should be built with the capacity to train and function according to their anticipated usage. But as machine-learning sophistications grows the demand for compute does too, the question becomes whether compute should or even could keep up with demand, or whether there are more efficient ways to achieve the same results.\n\nThis paper examines developments in architectures and approaches in four cyber-physical compute areas.\n\n1. Advancements in harnessing FOG computing to enhance computational capabilities closer to the edge while maintaining low latency.\n2. Developments in telecommunications, and its significance in Machine Learning.\n3. Developments in Big Data processing and the connection to telecommunications and FOG.\n4. Developments in Edge Computing: Improving on a Federated Learning approach.","metadata":{}},{"cell_type":"markdown","source":"## What are Cyber-physical systems (CPS)?\nAs a point of definition, Cyber-physical systems (CPS) are systems in which physical actions are controlled and driven by software. In CPS, the physical world and the virtual world interact and influence each other, with software algorithms playing a key role in controlling and orchestrating the behaviour of physical components. [6] This integration allows for advanced functionalities and automation in various domains such as robotics, smart homes, smart cities, autonomous vehicles, and industrial control systems.","metadata":{}},{"cell_type":"markdown","source":"## The Target Future State: Driving Decisions Through Actionable Insights in Cyber-Physical Systems\n\nMaking reliable decisions and taking effective actions in physical systems involves mastering the skill of condensing a vast amount of information into a single well-timed and contextually significant insight. \n\nThe complexity of the challenge can be understood through a simple scenario of a human crossing a street. Success in this situation requires the interpretation of a significant amount of sensory data (visual and auditory) - essentially, understanding the environment at a given moment. Consider the multitude of elements in this environment: the composition of the road, the manufacturing process of that material, its physical properties, the road's width, potential hazards like poles or fences, adjacent features like lawns and houses and all their attributes and associated processes, the current temperature, the road's slipperiness relative to weather conditions, visibility, traffic conditions, and so forth. The list is practically endless, illustrating the vast complexity of a seemingly simple decision.\n\nIn this paper, the concept that an abundance of information can be streamlined to support a simple decision-making process is termed \"actionable insight.\n\nThe hypothesis is that actionable insights will drive the next generation of cyber-physical systems.\n\nThe issue at hand is that, in the majority of scenarios, these actionable insights demand substantial computational resources.\n\nIn the context of cyber-physical systems (CPS), a significant challenge arises from monitoring large and complex environments, in the machine learning world referred to as ‘state spaces’. The exploration of these state spaces is a fundamental aspect of reinforcement learning, which is frequently employed in CPS. However, the computational demands associated with exploring such extensive state spaces can become prohibitive.\n\nTo address this challenge, various industry bodies have tackled specific aspects of the problem, resulting in the emergence of hybrid architectures and standards aimed at improving the processing capacity for machines that exist in dynamic and information rich state spaces, AND, this last part is key - ‘within a reasonable timeframe’.\n","metadata":{}},{"cell_type":"markdown","source":"# 2022 - 2023 Developments Supporting Massive CPS Compute\n\n## Development #1. Multi-Agent Models and Fog Computing used to drive Low-latency use cases\n\n### 1.1 What is FOG compute?\n\nCoined by Cisco in 2012, the term draws an analogy from the physical world where fog is closer to the ground than clouds in the sky. Similarly, fog computing brings computing resources closer to the target devices, resulting in reduced latency and improved responsiveness. In 2021, Nvidia joined AWS, Cisco, Microsoft, and Intel in providing hardware and software solutions capable of operating as fog nodes, enabling the execution of machine learning software within this decentralized computing paradigm.\n\n![image.png](attachment:f4f466c7-4a79-425f-bd0d-0fc20220486a.png)!\n\n    (Source: Embedded Computing Design)\n\n#### 1.1.1 Sample Use Case\n\nConsider a robot designed to function in an open environment, which may utilize more than 20 sensors. Some of these sensors, like a proximity sensor, provide straightforward numeric values. Others, like a camera, yield more abstract data such as images. Further, some sensors deliver data that requires context for interpretation, such as overlaying GPS coordinates onto a GIS map. The computational power needed to convert this diverse and complex data into actionable insights in a deep-learning model is considerable.\n\nNow, imagine the challenge of synchronizing 100 of these robots to work towards a common objective. The computational requirements for this scale of operation are significantly higher, demonstrating a simple example where FOG computing with its low latency and compute capability may be better suited than 100 individual machines trying to interpret the world and coordinate at the same time.\n\nWith its low latency and robust computation capabilities, FOG computing could be more effective than having each of the 100 individual machines attempt to interpret their environment and synchronize their actions independently.\n\n#### 1.1.2 Commercial Considerations\n\nFor the purposes of this discussion, four reasonable computing options in the realm of cyber-physical systems are considered. To provide a sense of scale, current prices are provided in Australian dollars, although these are only indicative. The economic incentives to develop advanced solutions that utilize lower-end computing resources whenever feasible are clearly apparent.\n\nIn an environment where the benefits of actionable insights are clear, but the need to keep individual unit costs low is imperative, FOG computing emerges as a viable alternative.\n\n\n| Compute Architecture | Example Hardware | Features | Power Consumption | Indicative Cost AUD |\n| -------------------- | ---------------- | -------- | ----------------- | ---- |\n| Microcontrollers| Raspberry Pi Pico| Basic| Very Low | \\$10 |\n| RISC Computing| Raspberry Pi 4 | Small instruction set| Low | \\$130 |\n| CPU | Intel Core i7 CPU | Full instruction set | High power consumption | \\$500 |\n| GPU | Nvidia GeForce RTX 3090 | GPU instruction set | High power consumption | \\$1900 |\n<center> Table 1. Indicative price comparison of compute architectures </center>\n\n\n### 1.2 Multi-Agent Use Case\nMulti-agent reinforcement learning and FOG computing are well suited because the amalgamation of diverse perspectives (from multiple agents) can accelerate the learning process. Moreover, the aggregate computational capacity in a FOG environment surpasses that of any single edge computing system, potentially enhancing overall performance. Furthermore, offloading computation to the FOG layer lessens the load on the edge nodes. As a result, these nodes could be equipped with less expensive, and thus more cost-effective, computational architectures.\n\nOne example of a Multi-Agent Fog computing solution is outlined in the paper: MAFC: Multi-Agent Fog Computing Model for Healthcare Critical Tasks Management evaluates ECG monitoring in healthcare. “ In time-critical services such as healthcare, Fog computing is perceived to be cost-effective compared to Cloud computing because of its lower latency and, in some cases, because of the spare capacity of locally available resources” [7]. \n\nThe paper expands on the idea of utilizing a Multi-agent model within a FOG computing environment to address time-critical tasks. While the healthcare industry serves as a relevant case study, the proposed architecture has the potential to be applied in various industries beyond healthcare, highlighting its versatility and broad applicability. This innovative approach of combining Multi-agent modeling with FOG computing opens up possibilities for optimizing time-sensitive tasks across different domains.\n\nIn the paper \"Multi-agent and reinforcement learning based code offloading in mobile fog\" [23] The authors propose a distributed reinforcement learning algorithm to offload basic blocks in a decentralized fashion to deploy mobile codes on geographically distributed mobile fogs. In this example the distributed algorithm runs across multiple FOG nodes, further increasing compute capability and reducing burden on edge nodes.\n\n### 1.3 FOG Conclusion\n\nFOG computing emerges as a promising commercial alternative for robotics that need to scale, cooperate, react in a timely manner and leverage sophisticated models. A combination of developments in infrastructure based solutions (also known as the fog layer) and reinforcement learning algorithms, particularly in the form of multi-agent environments offer significant potential in low cost but sophisticated cyber-physical solutions.\n\n## Development #2. 5G-V2X Standards Present New Opportunities in Reinforcement Learning\n\nAn important point of order related to this section. 5G-V2X is a variant of C-V2X (Cellular Vehicle-to-Everything) that operates over the 5G network infrastructure. C-V2X operates over existing 4G infrastructure. This means that the potential benefits and capabilities discussed in this section are applicable today even without widespread availability or maturity of 5G tower rollouts. C-V2X can still provide significant advantages and functionality using the existing 4G network. Where possible, a reference to ‘V2X’ generally denotes the use of the protocol itself, without specifying any particular variant.\n\n### 2.1 Why is 5G-V2X discussed in a Machine Learning Essay?\n\nA key capability that is often underrepresented in large-scale commercial operations is wireless coveage in wide geographical areas. Whilst solutions exist today through telecommunications carriers the complexity of connecting these solutions into corporate industrial networks is significant.\n\nWhilst most mature organisations have established data science capabilities, connecting input from the field to a machine learning model is often a challenge. Particularly in remote locations which are representative of many energy, mining and agricultural sites. The developments listed below related to V2X and Robot Operating System represent a conduit for the application of machine learning models where solutions were evident, but the logistics of communications are a constraint.\n\nIt is important to note that the application of 5G-V2X extends well beyond self-driving vehicles. The fast re-coordination of resources in complex industrial settings such as mines, for example, is a significant challenge when there is an unscheduled vehicle breakdown. This is just one instance where V2X technologies and reiforcement learning could provide significant benefits. Hence, its inclusion as a significant development in the delivery of actionable insights in Cyber-physical systems.\n\n### 2.2 What are the 5G and V2X developments of interest?\n\n3GPP, the mobile telecommunications standards body have released updates designed to enable cyber-physical systems to operate under Cellular and 5G coverage. \n\n“In July 2020, 3GPP Release 16 introduced additional capability to 5G, specifically in terms of short-range direct communication, increasing bandwidth and reducing latency further and is sometimes referred to as Ultra Reliable Low Latency Communications (URLLC). 5G-V2X offers the features which are paramount to highly and fully automated and cooperative driving such as the exchange of: \n\n•\tSensor data sharing for collective perception (e.g., video data)  \n•\tControl information for platoons from very close driving vehicles (only a few meters gap)\n•\tVehicle trajectories to prevent collisions (cooperative decision making)” [1]\n\nThe application of V2X (Vehicle-to-Everything) to Robotics via ROS (Robot Operating System) provides a key communications conduit to inform Machine Learning. Most notably Reinforcement Learning and Computer Vision systems.\n\n### 2.3 Robot Operating System Platform Support for V2X\n\nIn 2019, the ROS2 (Robot Operating System 2) V2X Working Group introduced several V2X packages for the platform, thereby enabling V2X messages to be conveyed. In 2023, ROS 2 released communications using WAVE 1.1 (Wireless Access in Vehicular Environments), a suite of protocols designed for wireless communication among vehicles (V2V), vehicles and infrastructure (V2I), and vehicles and pedestrians (V2P). This three-point communication framework serves as the conduit for real-time data, supplying machine learning systems with the data points needed to strategize and execute tasks. What sets it apart from Wi-Fi or local communications is the extensive reach offered by 5G and cellular systems. By leveraging this expansive infrastructure city-wide solutions are within reach.\n\n### 2.4 V2X and AI Research Developments Focus on Scaling Up\n\nUnsurprisingly, the use of V2X (Vehicle-to-Everything) communications to facilitate machine learning node communications has stimulated research on managing large state spaces. As these technologies are applied to more open environments, their inefficiencies become more apparent. In the process of adopting this innovative approach, it becomes both logical and necessary to continue refining our methods for handling these complex state spaces.\n\nRecently, more research papers have emerged that seek to better understand the correlation, role, and benefits, of Machine Learning with V2X. Unsurprisingly, there are various approaches proposed related to the size and complexity of the state space.\n\nIn the paper “Reinforcement Learning in Robotics: Applications and Real-World Challenges”, the authors outline a number of state-of-the-art algorithms and policy representations. In essence, a reinforcement learning system needs to engage in trial and error and adopt inefficient strategies to discover a range of possibilities. This exploration helps confirm the most effective solution over time. However, as the complexity of situations (state spaces) increases, the scalability of traditional reinforcement learning methods becomes a challenge. Novel algorithms, like those discussed in this paper, aim to find a solution in a more efficient manner by accelerating the convergence process.  “Traditional RL approaches based on MDP/POMDP/discretized state and action spaces have problems scaling up to work in robotics, because they suffer severely from the curse of dimensionality.” [5]\n\nIn the paper “Artificial Intelligence and Machine Learning as key enablers for V2X communications: A comprehensive survey”, the authors suggest that machine learning will play a key part in predictive decisions and navigating the complexity of the state space. [2]. \n\nA Real world application of machine learning to an open state space is outlined in the research paper, “Multi-robot path planning based on a deep reinforcement learning DQN algorithm”. Here, the authors evaluate a Deep Q-Network algorithm in a variety of scenarios. [3]\n\nIn the paper “Resource Allocation in V2X Communications Based on Multi-Agent Reinforcement Learning with Attention Mechanism”. The authors propose a framework of multi-agent deep reinforcement learning based on attention mechanism (AMARL) [4] to improve on the dynamic allocation of resources such as bandwidth, spectrum and time-slots in a Cellular C-V2X environments. \n\n### 2.5 5G-V2X Conclusion\n\nUtilizing 5G infrastructure, which is already propelled by substantial commercial interests, creates a significant new avenue for deploying machine learning in massive environments. The terms \"vehicle-to-machine\", \"vehicle-to-infrastructure\", and \"vehicle-to-pedestrians\" don't fully capture the scope of this technology, as its application goes well beyond vehicles alone and the potential for innovation is massive.\n\n## Development #3. Computation Efficiencies in Big Data Enabling Adaptive Robotics\n\n### 3.1 How is Big Data Computation connected to Cyber-physical systems?\n\nTraditionally, Big Data is used to provide insights into trends and patterns in unstructured data sets. Its success can be attributed to its capacity to transcend the limitations of predictive disciplines by moving beyond conventional structured databases. It enabled a profound exploration of entirely new and colossal repositories of unstructured data.\n\nCyber-physical systems stand to gain significantly from Big Data's ability to uncover insights from intricate relationships. Consider a typical robot equipped with a camera, GPS, accelerometer, gyroscope, proximity sensor, and Lidar. The diversity of data generated by this suite of sensors encapsulates the essence of Big Data. When analyzing such a large, varied dataset, the results could be revealing insights, classifications, or predictions. These outcomes align precisely with the valuable insights referenced in the introductory analogy of navigating through a forest: \"The path is safe, continue walking,\" \"The ground looks unstable, tread carefully,\" \"The surface is slippery, take smaller steps,\" or \"Something feels off, stay alert.\"\n\nHowever, these benefits do not come without a price. The need for an alternative computation architecture, such as Fog computing, becomes evident in this context. Despite the clear advantages, the computational expenses associated with these sophisticated processes can be prohibitive. Development #3 has been included to outline some of the machine learning advancements that aim to mitigate the computational challenges present in cyber-physical systems.\n\n### 3.2 What is Big Data?\n\nBig Data refers to large and complex datasets that surpass the processing capabilities of traditional data processing systems. While Big Data is often associated with managing, storing, and analyzing massive volumes of data, its defining characteristic lies in the complexity and unstructured nature of the information. Unlike structured data found in tabular or relational databases, Big Data can include unstructured data types such as images or sound files. Analyzing and deriving meaningful patterns, trends, and insights from such complex and unstructured data requires significant computational resources and specialized techniques.\n\nOne example of a traditional candidate for the consumption of Big Data is Computer Vision, a core sense in many cyber-physical systems. In the 2013 paper “The Big Data methodology in computer vision systems”, the authors address the effect of compute restrictions on the quality of the outcomes. “Turning to the problems of developing computer vision systems (CVS), it should be noted that the traditional approach of using sequential processing steps performed by the separate frames of video data largely exhausted the resources to increase processing accuracy and the ability to adapt as conditions of surveillance change” [10]\n\nBut to more accurately outline the progression of this technology and its role in addressing these computational challenges, it's crucial to also discuss concepts like Fast Data and Actionable Data, as detailed below.\n\n### 3.3 The Emergence of Fast Data\n\nIn 2012, the quest to address this limitation gave rise to the concept of Fast Data, which emphasizes the timely processing and analysis of streaming data or data in motion. Fast Data technologies enable organizations to extract value from data in near real-time or with minimal delay, allowing for rapid decision-making and immediate actions based on up-to-date information.\n\nThis served as a crucial milestone for the implementation of real-time insights.\n\n### 3.4 Actionable Insights\n\nWhile Fast Data addresses the need for timely data processing, it is essential to transform that data into actionable insights. Actionable Data refers to processed and analyzed data that is specifically relevant and meaningful for decision-making and action-taking. Actionable Data focuses on extracting key insights, relevant patterns, and contextual information from the data, enabling organizations to make informed decisions and execute specific actions based on the derived insights.\n\nThe advancement from Big Data to Fast Data and Actionable Data signifies the shift from managing and analyzing large volumes of data to processing data quickly, extracting relevant insights, and facilitating real-time decision-making and action. This evolution enables organizations to derive value from data more efficiently, respond rapidly to changing circumstances, and drive more impactful outcomes. [9]\n\n### 3.5 2022 Developments - Performance Improvements in Complex State Spaces\n\nComputation efficiencies also remain at the forefront of Cyber-physical research. In 2022, two vision-based, multi-object grasp-pose estimation models, the MOGPE Real-Time (RT) and the MOGPE HighPrecision (HP) have yielded an 80% and a 96.67% success\nrate in a real-world robotic pick-and-place experiment. [12]. The combination of data augmentation and domain randomization were shown to improve performance of Sim2Real grasp pose estimation approaches. \n\nWhile the authors showcase positive outcomes in CPS training, they also emphasize the significant size of the state space involved. “Adaptive robotics plays an essential role in achieving truly co-creative cyber physical systems. In robotic manipulation tasks, one of the biggest challenges is to estimate the pose of given workpieces. Even though the recent deep-learning-based models show promising results, they require an immense dataset for training.” [12] \n\n### 3.6 Big Data Conclusion\n\nApart from exploring architectures that can link cyber-physical systems with low-latency computation, advancements in Gaussian process models showcase an example of increased efficiency in predictive analysis specifically applied to adaptive robotics. The key point from this section is that task execution, such as manipulative tasks, within cyber-physical systems are currently limited by the large state space they occupy. The need for continued algorithmic advancements in this area is paramount and holds substantial potential.\n\n## Development #4. Compute Improvements at the Edge\n\nEdge computing, characterized by endpoint devices equipped with computational abilities, is conceptually the closest model to our own human experience of having one body and one brain. This stands in contrast to FOG computing, where the device contains a smaller, less powerful \"brain\" or computational unit, while a second, more potent computational unit is located near the device. Cloud computing follows a similar structure to fog computing. The primary distinction, however, lies in the location of the second, stronger computational unit—the cloud—which is often physically distant.\n\nThe appeal of edge computing lies in its ability to consolidate all systems into a single entity. However, as observed in the fog computing section, the cost of computational power can be significantly high. The next section details advancements in edge computing aimed at enhancing computational capacity without necessarily escalating the hardware expenses. As you'll see, the overarching idea is to share computational resources between nodes to boost the power of a single model.\n\n### 4.1 Computation Offloading at the Edge\nEdge Computing, a natural fit for Cyber-Physical Systems, is both the benefactor and initiator of all the above-mentioned architectures and solutions. Ultimately, it is the need to conserve power and reduce costs whist producing actionable outcomes in CPS that imposes compute challenges outlined in this essay. Giants such as Intel, Nvidia, ARM and Qualcomm are tackling the edge compute challenges at the hardware level. However, the potential offered by high end compute, either in the fog layer or in the cloud, continues to drive research. The research paper ‘Computation offloading in mobile edge computing networks: A survey’ outlines the problem space “Computation offloading is one of the key technologies in Mobile Edge Computing (MEC), which makes up for the deficiencies of mobile devices in terms of storage resource, computing capacity, and energy efficiency.” [13]\n\n### 4.2 The Emergence of Federated Learning\nNoteworthy ideas at the Edge Layer include Federated learning, coined by Google in 2016 and Edge AI, coined by IBM that same year.\n\n### 4.3 Improvements in Reward systems boost Federated Learning\nIn the realm of Federated Learning, attention has been given to improving performance by quantifying and rewarding individual node contributions. Federated Learning is a decentralized machine learning approach where models are trained on local devices or servers without transferring raw data to a central location. Instead, only model updates are shared, allowing for privacy-preserving collaboration and improved scalability in distributed learning systems. [18]\n\nIn their paper, “A Hierarchical Federated Learning Incentive Mechanism in UAV-assisted Edge Computing Environment” [16] The authors propose a novel idea that a federated systems of devices may be structured into a layered hierarchy that is consistent with their compute capacity. A reward system incentivizes devices as a means to promote efficiencies in federated learning clusters. The more a node contributes, the more it is rewarded.\n\nThe authors of \"FedDNA: Federated learning using dynamic node alignment\" [17] propose a different approach for more efficient Federated Learning. They suggest that accuracy can be enhanced by pairing devices that have similar data distributions, enabling them to share more relevant updates with each other. This collaboration leads to improved accuracy in the learning process.\n\n### 4.4 Developments at the Edge - Novel Framework Shows Promise\nIn the realm of Edge AI, a subset of Edge Computing, the combination of Machine Learning with Statistical analysis has shown promise. Experiments on two real-world datasets from China and Belgium show that the proposed framework can obtain satisfactory prediction accuracy compared to existing approaches. [19]. With the rise of Renewable Energy, the Energy Industry is grappling with challenges introduced by the dynamic nature of renewable generation. The concept of Smart Microgrids is at the forefront, and the paper \"An Edge-AI Based Forecasting Approach for Improving Smart Microgrid Efficiency\". It explores the use of Machine Learning to extract prominent features, followed by applying statistical methods to these features for more accurate predictions. A similar approach can also be employed in Ensemble learning, where multiple predictions are combined in various ways to enhance prediction accuracy.\n\n### 4.5 Edge Compute Conclusion\n\nFederated learning presents significant potential in the deployment of larger machine learning models to low cost edge devices. There is a strong focus on finding efficiencies in the delivery of those models, for example by rewarding high contributing nodes, or by classifying nodes into hierarchies. The microgrids example is an excellent adaptation to its environment just like regular grids forecast based on their parameters.\n\nFederated learning holds considerable promise for deploying extensive machine learning models on affordable edge devices. There is a strong focus on finding efficiencies in the delivery of these models, such as incentivizing nodes that contribute significantly, or by organizing nodes into hierarchical structures. The microgrids scenario serves as a stellar prosecution of this technology, mirroring how regular grids forecast according to their parameters in a smaller scale with smaller compute.","metadata":{},"attachments":{"f4f466c7-4a79-425f-bd0d-0fc20220486a.png":{"image/png":"iVBORw0KGgoAAAANSUhEUgAAAUAAAADICAIAAAAWZq/8AAAgAElEQVR4nOy9abBcx3Um+J2TmffeqrfjPZAACJAAd4CkRYkiTUm2ZVuWR922LLemZY8lu7vdPZ5x/5gl5tdMTzja82siJmI8DndYdsSEY9ru9ow9tty2W5RMj5paaUmkSFncRHHBDmJ/wFuq6t6bmefMj7xVr94GAiSWB+B9UVGv3q275L2VX+bJs9Kh8yeVAHAZ/GK3c25h/tzC4mLVK3u1QE2WtVqtdmuUiOpeXZZlVfmyLC2bmZmZmaktzjlVVdVOpxNCiDEys80za21UiTGKB/UhUBEREVXtVqWIBIkxRlXlPgrjiCh9NsYYY5iZiIwx6dgQQtrfOWdtlmctZTbsyBmwrST0yrrn64VOt1P2Fhc6nU6n16uqqirLsg5+saqDCmlkEGmkqJYpY4JoxpQ5U7isnWftdnu01RrJswKGJUIjETk2WZZlzlprJ0ba1tpWnrVarXaRZVlmDTMjN5ZIgaW7Tp9xhaHNX1m+mYf/WfHd0FFLIARAgdRgu/zQwQl48G4u0KQ+mjNf8FEMdhsckt6Zl25h+NjB/hfG4JC0/+qrr9eeq/CTXRbQuXqRiIhtIK1DKEPdLesy+qoOMcYQhZmdy621pBBBCKHb7dZl5ZxrF0UimDGGiBpyEiXKKUGVGDz45URVVROBbeZEJGjapkvd3cfB/gMCYPkPMPRwWZSqyi90O/Pd7mKns9Dpznc7vbqa73S993UdYowSka4bVAIAJgY5w4bYEXKXFZmdHB0pMjfeHhkdabfzIsusszZjahlDCstkrc2sy7Isd8YYk1lnCNbazLIxzAQArEqQQfOGO8GV7hDrEZjW2Cftxqv3BsCQwbfL+b/20HDhuxomMC7iIQzTeHj76if5zgh88S25bghcVT0wga0ShBBUgkJUu1UdYvTexxiJuOmyxomI9957LyIkSkTcnzAHFFVVoYaBBmbwY8vQqFwHj/5jlaGuZfs7rx6/0/QbQvDel2VZlmXl67Oz53p11el1O72q8nUVJUj0MZKxAlVJxxpmZhARgY213MqLosiKPB9rtcbaIyPtYmxkNLPcdnnmjLXWMrs0NhGSRGCMySwzs2XDpNYaAxA1FKHUvTUaszQnXVLfvawYIpvy0EYGBIQhiq4DveC3w7jgbQ3m0jUHtdV7XmD7igMv5jwXM2NfL0RdD6QhgACwEgQaVZU4Qg0xAB0atlPHjApDSF+RAgABzcMhEEEVMf1HaE7Rhwz2BIYko6a7af8qNPTv8GNPnwcPXARRQRYRiBFCS70uTSLaP4/2eyv1z0sEKEjABMvNRQlgBQSqMARmEPXvZeid0zkURGAFSAkX6gXrCW9XEsMEXiUor2zI6qm1/xyHD1xxVP8Xf5speH0Grrcb1iHe9c60KwTS0Gfh0LsCIkCfZn0pBKKg/hYRqKoxtF7/0CHCrIYO/RzDPYiG9l/vJxtm8oCo0u+rAsTl88twZxtsH9wxD30YvnoSDCJUl8mhScgEE/dP0p/ZBt9fQyyjnAxt4bX3AQAZyA/NDdEl3cLlud/LJa1ci+HyWoK0jiDqTy792x7cvgz9q8uHW9WVDBv8Bml7WtymMSAdO3hP85em+StJWM13zaNfsT8heiGTVq+AQElTo1VESYgMSAUMiIIjlIH+XS3NnMMjxVK/U40xMohYiSx0aC9qhjMCK4SGztF0uFUrKx0sDq9JH1qDwKsItsaAK0vPpmHv6tVxAq/6yq694yXi8hL4XZ7kOgKpDE95wCq5qT+kKZRBAlXp65yanYaJugIDkjfHDr03XaF5VyzbMiDMeu+CCCFAmPpHDZ0zXWNVYwQEHQi8Qwuz/g5RlUjSiEZL4xrQnFlpqKOvp3+9RjPwmpIONSLJcJto9Z5Ltyn9AehiCCwAaB0C6+rHu4krAJKB3g9KfVkKQJRomVMXJsjwwJb0EmmIiyrcX/sqgdHMpAIlBRGtFmlWjLXD3Sm1YfjfpT1Bir6+enhq1f4id0DggYSt/RMM96QlCX/pbEPXXTGsRFo9EDS7rYKqArQusa8w1iFwXCU58QVWtgCAlYPruu+JwGsPWGuKshuN1RutPe8AFJYLgcMqpyVbzbJD+KJ/4OY8WCE+L8d6Q/3FgJfW58Mz8Cos77IXtD+svI91ft6VvVbX5sJVRVLaM/OSTUmFmAdPhBRpuMKwHoEgCtEIZgMCEAIsN8q6KGoNJYVleiJKIEVU8CrdXbLVJXvEVbrnmx4rCTwEwSqVDwBdZ8Rd9wL9U63GRRny3gZL1sjlq+bBejXNycOi+7LD+2L5ElYMQE37lwaIt7mLa9tzVTU5xhCRNRaEIJ5ImexS24SWLXmadZMAqH1gzZwFCDEIIMYOScjDCpFGLaJRJflaDJM2hGDt5Vkbb+LCWBKhlzYt+28V95RX77Q2LpKg767L69vM/6tOv5pwy2fPNNukeWbNtq15W+kx8dUg8OBSK7bqCmm//09omqwKVXD/F0zqCYXGGFSIrbU2ycUahIyAAUpGNA5lbfORJS0/FLb5BGrMTskJL9E4xmX28E1cOZCKXkjAHZ6yFE2/GZhlLoyL3O3dQWltAies6OZLZq2Lb9tqXdiqXYYvd00IrECUOFh2MnPf3odQl8wwhkHUKAgUUAUzwFCATVL7SYQxhAgiBUWF1+DJGiILsO9FlxdAv5+wQAVQUOOoM/C64TXVmZu4Mliuhb68uDrrQlptclp3klxp6b0YGl/w2wspdK8QNA5dl9G3qNPy5YEAyUvVMfeNg9InsADwITJbNg5EUeEjmGAMJMIYQINQtEBACHWwWctopgROXjoCITUUU3PSwjtNuclPdnP6vWq4CAKv6YKz3rcXs88VGzHWvOIabVmTdmsOAxdBx0vZ93JgHQL7GJmZiQWqIPTn6BgbTVXaSgRLEEAURAiKIAAhab58QGZRe2QOAMq6284KgL3AMaIiBhCBDYDG0y5HY0hP/vCb8/BVxioReq2uPzy0m/6GNb8dUvwMTvIOlF6XAl21Sl+xCl7jjlY7D65tbV5x2FrNXfPqlwHr+u4uf/gAtHF6bRwyEqSRq2G4+SwKGrLLOYYHFjs4+tbpN9489Nbx0z6i1WqNjrZu37lt7747RluoIpzBYlffeP3QwQNvdRbLsgzW2omJke07tu2587ZdWy15bVkQUVKeDWi8qYi+OqDVHWXF/2sFoch632KN/nupZqdLaT0Sgdc5mtZQuSnk4q9Iw2v+tfe4fATW/gk1EbJvn18yANDg39W/0UCZp0Av4NjRM9//wWsHDh8/caZTByhiUWQ7tm29887dd99z56232E4H/99T3/76N77Z7fn26ATBBVHDNsQqRh987+FHfuh9D7/n71/43tPf+Pbk5LThHGqMcURUVb3aV62Wm5nMf+U//wd3bJucHLEGiEEsE4hEhTnp9gd2uM0J+cJ4h664axB4BS5VHF498K4nnF66uLpO+/qd/u2bsqo9GwbS6Hip70VGCEHYGgGJKADDVEaQQakIHui7ZFgHw/AKEHzAd55/84kn/tPpU/NjE1sDuYodbG4MiYaq2/Ghbhf52NhYt9tVYgIrGwUTmTQ2xBhdZuuyd35+jiDjE1N5nnsfmZaiyvpjijCqavHEjz324Kd/7sMjjFGL0CttKwOl1TYFqTO2BKiaKGTMhnrmlx9pKZGM4UhiFCW1Yd9qrtAIIoiALUIUa5MbYhqE07pklZfB+hLN2xN4E1cUS3q0RODGNgOAI6QKYMcKVBHH3zr3wvdff+7lH8x3egvzvV6vkhCNMa123m4Xd969Z+cdt3//lde/99IPbtl6B9Cene1O3rK9F0MZvao455wh76u6rEKsW61W8tymxmTUXBX9LljXNYAsy2BMMhENtXlgfvd1ecb6hU/+1Ad//qMPmRptBzIIiqhCpBkpmsUIp6NuYAIn3lZVlec5gF6v12q1fKzZWAYDkKiGhyQoWiJwVfdcljEISqud+TYJvEGhQIQI2ABmuX2rFwI7G4G5Cs+98OaXv/bt/fvfUtcenZ6pVaGG2SbnUkNiDHXKjkDPn5+vgk5v2dluT9XBEJhsDKGGirXWkKaQ6oGiOPUVaeI+iQBu/LkoMVZE+l6zS60b8p+Jk2PuzImDunD2/jt37Nmx5SMf/tDWmZFWDgOQIGNB8DBGQUIGV8nSdm2QaJacWJYmYUChImLYJFMbRNmYxg5PiNFHhbNGVJgYSpekPtgk8LWEAh6iicDJxKUAQQkemO3EGubf/J9/fPCt85xNuWIKxnWqMhITJQKTSCSNRJFYS1+xsdblQVwIhpAxKcMzRSZSFQ1REbMsy3PX6/UAJA1FImQyR1jDGqIQEsNjjECTzwirCAyIhqqwqtV8bqP4hao7d++dd/4Xn/r5O7dnBWCjINZwNkYRm9FQvPeNh6TGG7igqWpVVXlRiAyH5aa1hQKQGEWJmWNU61zjaNPPx3SRF90k8LWEAh4AwMszGUSgFJxe0P/1f/+9heAkm1I3MdeR1sg4I4oGFVJtFNIEIVLiYDMTVSofiSxxBnWq0ZGq1ACIwSDVmKjY7yKN8KyAKgFieMkUlDTbKfJsDWUngRQMw4hABanygn3Z0dCjavFf/MLP/8Sjd2Q1ICWcVQnBFQq4G5fAadZNlvD0wRjT6/pWy9U+ZJmV6EOoszyTGDmNjyEam6lSr1e1W0WMyUR3CTr8TYfVawzTTLppfhMFRZgIBODf/tGflzHnbKIXi6AmH5vo+cDRWwKzsdYyjKpqDIpAhM5Cx2Yuc7YqPTtkztZ1SJ5SqhGS0v8kwTgsS/0DoKG0EYnMA8Y24pyIrNmllDgKorCSsy471+u4bKzdnpCFs3/9N0/dt+uXd884qEHwlDl6d4ErGxzD3qMxxuQP99Wvfv3Y0ePOOZeZn/mZf+CcyUze63Za7XYMoaqq9sior2uQK/ICwDuYTDeV+9cSBBiFVRgIwUfUET5CI3DidHzlB0eYxp0bJ3IAiIW5bhecO3KIqCspe1qVrMGo2CjtLHeIGWSi7ThW5cLZloVErwRmq6q190EiGc6KlhIrsaYIpURyglJjMVYQiFenj2v2oSXjFhHZzIVIHta2twRud2sjrn3kxOwbB48opywRulYU8g2FtOLt9XrMnGUZMz/55JNf+MLnP/HzH3/fIw9/9atfHWSAaLVH6qo21rXbYxLVuSKE0P+ybzu8aCpvzsDXFDqY/liBCAowERSA03MhoAXTXuwGl7fYkA9dlhBESWFgKU2UDceSPwcLGV/Xopq7vHBU1TWxaRyz2KbIQpGYpoiVqR4JA9kviX8YmkzW6VJCRIJocytKqqbbDePtQtSPTs0cPzfngcw2SU5iDGxu8P7WarXSCiWE8Kd/+qe/8iu/MjLSuv322z/zmc8A+J3f/jfTM1MHDx781Kc+9T//q9/40Ic+dPbs2U996heffe47x986+clPfvK++++51CtuzsDXGiE5bXA3Go/szVNzf/Glb/1vv/9Xn/3DP6m41QvsspYPNYlnLS0HsKqhSIiEwPCgAIrEnq0nE8kpF2RaQdhHIjYAN5kQVAUqUBATG6TpN02nfe0oKQTJtJSCFZQMJ3E6Le0oTdlDL6Uo4kk8S+QQRotWDCGIdqrq7n17a0bV1+Ew5AZWQQ/WGswcYwQwOzvbbrehgMT3v/993/72tw8cOvhLn/4nrfb4333z2dbI6Cf/8af+y//qv/7d3/ssgPZI9ld/+efcH9E318DXDyy6HTFtPj0X/vSvn3jmxVdb47d0vbX5pPqgzqlhEgKEUkqg5NlGuizOuRmIeSi71TvzbVu/mdYu0ZgoTcvECFJBlWEAUYlBYl3XEnpa+tcPHdm97Z7JnH2sVWLmWiEE3LhxwoMoaOdcXdf33nvv9773vQf23T82Pv7yyy977/M8J8L4+PiZM2dUdXJyst1unzp16qd+6idbrXx0ZEQ0MNwlXdT85m/+5pW5nU1cBAgxwhZ0ah7/x2f/8NWj8/nUHbNda1ozatsRzM6BETWAWJOimqApEyCRkjb+ZE2uDfRnOGrCGUgb00Ua2tcZ1qkZ+Jf2oRT5MDRhplmFiIwxhomAGIMPNRslglExAIlqiDEGDX607Y4dfvP2226dnhozhMLa2lfO3cBKaDBzWZbW2qqqnHMPPvjgb//2byv0pZdeUpWHH37PV772FWJ++u++8enP/NIf/99/vOv2nffdd29rJP/jf//vs9yeO39u1+27iGigO7wYbJqRrjFE0Q343BNP/81Xn6fx7d5MmNZEtw5CGn0vY7CBhEhkCCxQUFyaXZWBRNLk6jTkmkrSVxstU/3qWh1jhYJJaQ2VUzIFN+k+oNyswFVYCGKisoLECsirKuqWC+Xc0X27p/7bX/vMTA5IZdlIjGzcjbpwK8uyKIrBPFxVFREdPHhwfHx8amoCTFXVOz93bsuWLSHUWZZVVTUxPhUlzp45G2Ocnp7OsjzJU5si9PUBBZRQ1vWps2fGJ6fmgvNKVcfXotaBOZl/ACYRMFmoEpR0EEQxCO7XJveYMkiSY2Y/+dcyJ6qUbGSAQWr+oX14iL0y+FZiBGDNkmqaFQKCsiAyRADDRGwyYpARvzAzvX3/G6+Xvai5kaiWISJ8g0YKxxiLohCRtNYAkBwq773nXjS6/Tpz7daoDVK1WtnswuyW8S2KEkxT0+PO5ASjzVh8Cbgxx8LrCjI2kk1Njp6dPdluOYreQCdG2kY1J8MqyVlClYQYYBIDMIuFMi9jWuLiIOvt4LvmwzvKPrKEEEJaAKd5OMaYvCxFCOoENhX0iCI+ablhfSXMmXiViNwVUaJ1l7bAu+4QQkgfUiUgACIoexEAE1XaZXjD9WuHvvsn/+8fvLr/u6XMAd4ZFogP/h2kNN0k8LUEQYIvo4af+7mP3DozPnf26ESbrfZ6c6edLwuojWqjWm2y70aowiqcwCiMwCqsEiuMgpRYwc17UhITALDqkMFqTSigvFZ2tAGli6Kw1ooPvcXOwvm57sKi956ZlZzAgqySiYTE4SAQQVWGe+64b3LcWkZVecNG1gjevkGQrG7W2mR4G2wnhnXJ4TwYUo/F1w6+9Pkvfq41yn/1xJ8ePvK6164iMpjYBomXfN1NJdY1hELJqBCRMe995JFOKW+8eXBuYaGVF6rRWBNCUCZiE1WYjED6uioa8r/kZtnbiOQKEIgIqfqaEknjFNnPAzusoOpL2kKkaTlN4MZTY0iYLquKmLM8s84xKMZY13WvqlrFGEAMgQojMkXHmiEWFKq50//ilz+1c6vxFVq5iRKZzDXLm33lkZyxkkdHXdfWWu1nO6mkMhyA7uuHX3jiyT/bdtuEzTExNfL83z8/MT46Mz0DWKIc0uQ/2lwDXzcIiAQW1clR+tEPPfq1b35nbGTy/MLpAOPP6+j42Ggx6mMgYmNVQwRFBUgMYEkMYEAKCCOCRAah0WpJHCBMESQMDIVML4UiNH8pDFJMGlgVNsZE8VAok0gEE1tjnI2qIYrNs3a7XVbdc3ML3e5Cuz0q0edGNVSFiRzq2JsjqX/9V/7RA3dnpHAZlAByN64GGsMxwDFG5xz6CsVaoOw9Oq8efvapr/7lxC3wbhag4DFzW+tvv/Y5WL9vzwe9p9yOCsQQAPXeD4YDAOsJy5sEvrYQC9MJZWbHFgL+3R/94ejoaDDFji23ephOt4QxsI4RNUqovYjw6lVkyqmztq9ymlgtpVhxZQCsLASGCBgkSiDlfqUpVhHVlCVahFItGVVRsrb0NRGxMyRIn6enJt46cXystZ3hEerJnM+feQu+88hD933iH370jh0jjhACMgs07oE3LIMHzmpEtBSQRFJ7bzMYhFcOv/CFv/mzmdtaxpVkVcBZnknVHdnCf/vl/8Amv+/2DxCiikRBql8/OHmM0ZhNAm88EMAq47boKeZO9U6fmM0nd1TRzJ3vmmI0Iich7xkhqqAwLmsX3apcytbFzZIpNoalQU0GVgAUlRApUwyJzpB+1i8WQnKBJsggDzabSJCgkYiINSqRgYKjRmIWgcZo2bZaLUQJoXPbdN49u9+QOI1+1H3kR977ocfeu+PW9oiDBUTgbBL7RRGA7AZWu4QQBuxN4R9EIc96izJ75Oj+//jEn2y7bZKzuqwqA66DFAVZayYms5E8+w9//W9/4efyO29/T87jEnng15Uity+Q5XOTwNcWTIJQl0Xhts20Ws5CxRHdMj0zu9ixJlPAMBetUZXge+XC+TnTakMBRJBAJS1T+xntkkuWcEo7SSpE0Aww2kjLCoJAlETAfXsSKzE0cV6UhAkSBUSqEFFrMyIDUREYw87ksSq7nU5mzMSInT198ic/9PC++++7Z8+OwmEkQ8ZwDA0qLJYBjUm8tMasl5X+xsBA4h0EbBJihbkDB1568kufv+OOWznrLdbzE1OTi53zrVarqiq2xve6Gsy+fXue+OKf/PRP6EP3fsgYl7T9A956753L17zoJoGvNcgaiEQUDp/42f/s3//5E9nEtvkzPZe3iTQqfBm6MJm11rLL2pVGEIAAikqN46QQFHZQF1kQEo1ZLdT2/TsEJILIJELCS64g0p+6GYpISsaqijEWQiF6ZU72XgZrVO99ZlxrzFEMoTc7kfl//LEfvWXKplElM4CoIyI7cO2iqI1rWJRobtB0s01VqqHPIlLWi0dOvv6fnvriLbdOV3XXcslGemXHS2TViYmxcnGOSLfMbDlz4uSW6S1f/dqX4Ef33vPeLMuS61vyn3Hrm982CXytQUR5K02dH3j/vXmr+LO//hvvg3PjUYrC5lKYGETEE1nVyMoY9pRqlMiMfgJIVgHA2uTZamZoiJIAkSGkwhBKrtQQ0uHMmymWQaNXY4whUo0GpD4RT0QDq1pVCjFU3dg5/auf/JmZMRsrbedUx8gEy6Tio4iIOGdBZNimJJsEc6NOv+jHACbiASjL8tzc3Of+/C+VyhDOeOq0J+LIlJUYxsZG5s53FF5inZni4IHDnTkay1vdhfDFL35R6uyhhx6y1qZ19UA9tiY2XSmvMXxdgox1TlIijoBTs52XfvDGd1/8/tGTZ8/NlxF5GRDh2iMTeTGmlCtMU2GQIP0Y3UjNVNw3Cjfv2i/YnVwrlYQgw9okbgjMyWpMjBhjd2FxpFXkWea9z7KsrkvnXFX1rKNWbsrO+c7iue1btzxwz20femjPvrt2tnKua2FSZ1Pi8NBkkekn64kxWmtuqlzRIYSFxdmoi+xqcnJq7uDnn/yjiVtsFRazVqZKADIyocLief34x365ZW7NeFrrbHJsxjmXVtSDoE5j1p6EN2fgawybFWhkS6p8Vbh81y3tHbe85yd/9D1ziyGynV0I3/j2d5957qVeuVDY3AdROFKGshIpODJEAW7YS8qksAIjAElMUYJLaLJD9j0rG0sSQUAwKk6lrBbruTPOF1l7xCBmtqWxRJSxnOuyM3vm3Jap1o/9+Hs+8MOP3Lmj/cIzLxw+GO67705nmJIlOiobO0jeK6JsiG3KIIMbNSJ4qbbrkDZranIGmOr5eeOwdVoWF3sTt4yPj493yo4xLsYoRHk2empxsTBTt47f6b0tRtveRwDW2uQqkwK/1rvuhnucgwyaw6k0k3dL0sitrny3Ys/hxLwbHNr4QCozFLHtGPACtVAS3DraqoG5M7OPPbjnw48/8jdPfuk7333FFJOE3JDJsqL2GkSLolWpQlkFeZaF2hvFuVOnto6PZ62iIik1Gs5F1OVFDOp9TP66qlElMMNZQ+o1CqGiME/l6f/lf/pvFubmDrz5+uL8wvz8vLEZkY6OtXZsu/OO3Tt23nbL9FienEgeee9D3/q7b7ZGRm7bfmuydEiTrAvc+Iv0nbFvaFFvOEHCUGlVVuHCTUeUCM7ZFhFVVZVk45QIpa4kc2OGxhTt3BUAnGu67mDpewG/jg1H4AEGhXYGhbOSo6kxpq7rtMpfkYUoiRzXC3vR96LqxxykDOBCAIELNrXUdRVOHXnzoYfe3xrjf/qpn/7Ex37yS19/9tDh44cPH64Wo7N53hrtdc97WLKtsgqdqLEqW5m96/aZ7rkznXMnaWxsdGyiKnsxAtGIF1ayhjRqFGHLEB+qxdwxIZaLJ0eL8D/89//87t2uLGd+/JGZOkAERLAW6CcdNwAjKLwIyLU+9KMffOqpr7RH21OTY71e3W5lAmgAUcpV0A95pBt2+r0wSMHkoLYx9S3VBRn88gZqLzUSOGHDPdHh6XeQDzG9J6HCOZdlWbKzqWoIwTk3PCenyOlrdweXBhIlSj6OLE2SLANwFGWmk8eO7dh661TbWcJogS2F/eVPfqD0mF/onTt39sDhQyfPzs/Oe87Gn3r6u8StIm8XLey7a/s/+8xHJgr8xV99+RvPvFjW563J262JzuLpzOZFPlpV85aND5UVbbeNr6pTx9/aOjX+Yz/yno99+H07pvJeVbedg9e2YWMhihBB1NQXMCpMCrJgBKDj/aOPP/b333v+gX0PTU9N1Mmj0KYwKrAhQAbZpm9gJdaa6MdoC/omAwAAQ7lfhW6wxImES/Yz3XAETuuHYYFkhfyQMt+nuC0MiRlYbkm/PrBMqmRApZ/5nJiqSk8eP/2BH35UfTDGxrpUUJbl7NCezLdt2bb3ru0Rpgu8caj7vRdey1qTzhahWnzpu9/OPv2RjPBPPvUTH/voh599/pWXXn71xMnT9dz82PiWsnOayNQiMVZkTaiwfduWT370Ez+0787MYnoUEsrxPGPEiGDZ1pU3NsvZMA+ijRlRIVEJgXyR21bu7r777pdfefGRRx7JslSIICUNkaaYFqlq7Pe3m4rDw5Ekw3W8GIh9Yqe0pOEd8HHD9fVh+g2yQKTVb5KcAQw+DLLgA2DmwbGDoOrrCKqpKDcNfu0D+49MT2+1hthYQExmDCgSLMDMUTWQAhgFDr1+oLfQNTTWGqWiaMdWce7U2V33TIvKrna9SCkAACAASURBVEne9ZMPfvzHH5ybx2K3mj0/9/TfPfPggw8S0cyWqZF2MVq4LZPEBBZYixBiYQuCxCAUBZDMOrBRwSBUhgGwARkiOBKFCnTnrdtiXf39c995/PHHDUggBIoSDMgyEYhvXD/K9SGg2HzocxXA8lJeqU56ACIQcYmT8Ibr5UnWStQdLhudSNvpdIYF5sTSwYq3qqpBXPU1av4lIglYqprSqkMAViVVLC7UZ8+e/sDjj4jAWKgKaQRzrAMbQxBGLEzmgQPHe1/+T18RHzoL874qCxtHC3v3nmkJdcsSwF7UMtlx3DKZ79lxy+yprR98dLcxEI+Wa3J5EEAG0YfC2aoOBMkyC9N4TMdYG5M1a1rqzygEEAxx1OCIBdi54zZfhZdeePm+vfePFAUAYqcxiiAl0OqvAG9mLBO6hkpPDkeYXMIj2nBPMwnPzrlE1ERdEcmyLIQwMjIy2J4s2E3JD9W0SG63291uN83P1wWUEDjVJhSCGATWCrE+cezgzu1bMwsFKoGw7SmUOMtgTTRGLTsIosfr33/z5Fsn88wxB5HuzNaRX//1f1ZYxFBSOqeKBXJG5es6eINSxRugcFDVEDxBSQJ8baxVQZbZzGUSVBUwHKM3liLqgBhJGsszI2oMMWgUikRgEuTG3nnHnbnNDx04XHY9AwbEZAELMNSuXUTyRkaj9SMYWn7vQ3lR0qLJDgWHXgI23Aw8KAIQQjh48ODrr7/e6/VGR0dff/31F1988bOf/WxKHVYUKZP90niWciB0Op3f//3fr6rqN37jN67ZPVw0Bq2PUDNkPxOJJ44f+fGf+EganawhAGxcQHQgqIook2Egd5g7e2rExao6Z3P3a7/6zx/ct32qBRGMFu0QShV1rgjilantrADE3ppo4YDARMYSpXqDzFAFSAQaYZ2R6AlkkphDLFCFpALLBsQm5ZLm6D2LYUZdB2ftAw/s/bu/+9bZ0XFgut12jQZjWXqfmwnKfb+5ZcwkFQVIWRtiv0Mv8Q06IsYYjx8//lu/9Vt79+59/PHHn3jiib179xZFsVTzckjdFULw3id1dJ7nH/zgB1PCl9Wn3WhuZ9QvzpCRY7K1V8AZk3/3ey8/+NB76rpmhnOkEhnIiEgoiEQhNg5MKQbwsUfu/dAjd//cT7znf/z1X/zRH9q+rQWnKBgAs8lcloGEyECZkTJ7sCpZhVNjkkkoJaYlTnENhmEdAPQT0BnAEJjBBsbCWliG6RcNVeMsGAI1jqNqjPGxx96//9CBU2dPJ1tTUFVC5UP/6cvbvW4gJGsoNNW3YSVWwwBpZIiKMeREPRsRFW0WUDRkcOILrzs23AycQERPPvnkxMTE7t27e73er/3ar1VVVZZllmVPPPHEqVOnzp8//9hjj33uc5/71V/91e985zsAPv3pTz/77LOvvPLKkSNHer3esO564OlxSYUbrw5C7VOmqBgkc1kIstDtqdLoxLjLMyQ3j8ZpEibli1NJ3lQ+1JnL7r1z57bPfCrLsizLiDWEYABjHQCm5vdNnKDkG6UMZQKaDBzLckqvxlLXWfPZLdVYSadvKrTgkUfe+61vfWtkpDW9Zar2wbC1mY0S7Y2a1e5tMDwwUX/LUkSYrs4DenHYcDPwwHfsyJEj27dvr6qq1Wrt27dvenqaiE6cOPGFL3zhp3/6px999NHPf/7zIYRer7dr164jR45UVfW7v/u7H/3oRz/ykY9cIH5yo8FmrvZ18o4CEGN85ZWXbr99Z5FnjeyZCgUOvQ+EC2NMVVcx6vj4aFFk6StrrbNOoaIiIiscoC77EEYKBvVz9PQTvhO1WsXDDz/8wgsvzC8sXncWgesIG47AAKqq8t7ff//9P/jBD7z3qnrixInZ2VlrbVmWCwsLY2Njxpi5uTkAeZ4XRUFEZ8+eDSFs3bo1hJAMxQOBmZa8+TaWCJ1grXXOACjLutvtMvPMzEwqMhYlDlxh0x0p1BgTJUaJKW+DqnofQ5DkgkYgxZIaP11iRSWkK4cBh0V0y9Tk3r17n3322RCCDxEAsdmIP8D1jA03NKpqlmVE9IlPfOKpp576vd/7vUcfffT48eNjY2MnT54cHR3dtWvX17/+9QMHDnz0ox89evToE088kWXZwYMHb7nlln379v3BH/yBqr722muDcMprfUNvg+QQGqICMMa89tpr99x1V2YsBAQQMzUrZSUSZg4SDRvDrNCGLat+QwJZY5dktlQM/soglYpo6jkMtgGOEaNuu/WWuq6/+e1vPf74417EDYUoXRDDwvxGnGM2DjYigQEkS9Lv/M7vHDp0aGFh4eMf/3iMce/eve12+1//63997NixvXv37t69u6qqAwcOjI+PJwesf/kv/+Xs7OzMzMzP/uzPXuv7uFj0Kx6Ic/bcuTkA7XY76YOp74bX9wknDE+qyUNPJQYdeIBTyqvRp1KMKwPo6WopgkVgDNVRdu3a2el1X3rppQceeMBQk3VxE5cLG47Ag6ycxhhr7f3334++9Dg6OmqM8d7fdtttWZZ574ui2LdvX6r3Ya2dmJiYmpoioomJiUEs5TA2oBKrqkOW2SzjxcXuG2+88cADDzjnQhBjm8SvCkWjhEtZM4yIhFCnVEmGOAWK9olKqhpjE/VhjFMJjQKvT90r8hCGrERN+i2CKqzhsqrvv/fel155Zf/+/Xfsun20XayQopvqEBvvp7kusOHkk2QBcs4lNyzvPfpqZGNMWiKmUKSUtSQtklNFqVRYOf27rAL9hlz6JjhnAdR1SDdSFIVzZrgeL6FRCwFIVRHSE0izbtonDWHp0SVhpB8w+I61m+8Ey83yqGsPIM8zH+Ndd921uLh44sSJq9eamwMbjsCNKEiNuOicS9m90szcbrcTM00fzrmk5EwRSMMe0QNsnNF92DrdT30GL0rWPPv8c+973/uYOUpkhqF+OO3S0SkznBn+P30wxqQBbjiOsnmMRIn2AKJE9FfdaR9tYhgvD1Y859QYAgxxO8/vv/e+U6dOHT3ylgoRqK5C8EKgIBHLbX4bZ8AdOBSs51lwzbHhCJyiiwbPazhY/wYAESUuNQYhaAR6vd7Jkye3bdsWY2QmwyZeeomNC1ySjbHGpqun4ia9Xg9oTLdXblXM/eEkhKDA+Njo/fffv3///hMnTlSVz3PnnKkqb42t6mqovVdJYX5heO+TsDOIpRkWizYONhyBq6qq63pQ5hxDnf4GwMAm1NhLQQSMjLRff/313bt3t9sFESRGQ5f0u1zQhylGiXEwIojIyMgIM182jtDQa8VmatxEHBsNwsD05MRDD7/3zYOHSl9XIQLIcwcgc/mFazddfaRF3MDVb3jY3VDYcEqsPM+T4ffZZ59dWFjw3o+MjAyKvl2/SAJFknUHeYMFWtbCxhx/6+iRI0cMKRFplLRwWOdMw66zWPWZl38WFhUSD0pZJgV44dVXDx8+7LyQQlgAWAEAIdGVZ3g3YBHRviSfihsyc9Zqv/nmm88//53p6WmSJuWiMUYRgUG2zWtvRkph5/fcc88DDzzQarWuSRsuBhuOwGVZ5nm+devWvXv3Aogx5nleluXGt+heGKkTp3V7XdfNQpRJYL/z3HO/8Au/0O12i8yGEAyxqjZVepWx5K64Ioh0zc8rCWwUQlIrmJlUoyqybO/evW2YRGAiejcEXl0xPDWc2apq0swl33URAVEdZe/evc65559//qH3POC9Hx8ZLcvSZQZAP2ZYgH4SrWtUDC3P86qqJicnE3vTQLMBUzVtOAKnMCNr7e7du9NElEToG8Adb6A9GmQMFejBw2/t27dv586d1jADyVtYtV9QNIWqvI24e8H5ShQkHsTEKjGInDx3bteuXTPtEfSLfTfzHr2TgJjVMuWKxq7wIkmifFXHEML8/PnHf/gxCWJt3y9lxR2pXisCpxrITY2yfrrJDaiR2XAjCoAYY5qs0BfAbgD2DmNgo56fn3/l+99/z8MPGcOiKKuSiFKlz8t4MRClaHrDxlkXQsiyLNUMvowq6BWQpVouTfZNjRCBDyqAMWbf3nvHxiZeevlV770qmpCdYVxTdbQxJnkEYmO74m44AieNX0pLjyY4diOq7y8VgzjnlM1VgV5VHj95+u679oQQVOG9L/KCiauqetuzXdKVVSTEgL4ZKekIgUEA0RUBM0QwXNCbGMxwlnztrYEPevfdd1dVdeDwIdX+YqF/9BVq1SVhkBkG/X64AUXoDdegFSLKhbNab3Q0qtVGRcyMZPNUMkERQEeOHb17z525YUPaylxS5eZ5aykKdJV2dy1cMGpUlUiNIUU03FSRV9X0mJdrrpuc75cEWvVaOh2DTH9rk6sPKrGVOQZCVeaZuffeu0+fPnv42FFlBIEAdfAKKLj2npivYZzwsDPfhu2HG7FNNyTSSj7ZeJlZoPvfPLh7924iYhBfqWiDpU6vQ+/XEIaNqESJIyOtEGVsdOT9jz66/8Cho0ePNfEn1pVVKSqDAl+buAA2CXyVQGSYG28KInS73TNnztyx6/Zr3a6rjRhqQ0t1wHpl1crtXXfuPnz4cFVVZVWHKNZaJlZwPyXI22SluJmx+VCuKqLEGCMBRw8f2bFjh12n7PoNjCbNaIghhlbmjDEhyq6dt+3Zs+e5554TkRCCNXag0NqAeqMNhZuuA11VNIvCZvbw3hs2qnp+bv78+fP33L2n9te9g8qlIi0l8twxQ6GFs0QURXds33bXXXc988wzRZ550bIOAoSwlHL0Wjd8g2KTwFcJAx2mtXb//v3bt29HcsziDWdavKKQvjHcEEM0SnTcRAnfeuutd9xxx7efebaqqjxzVe3JNnWOsTkVr4NNAl8lRBVjbV2Hc7Nz3vvbb789hGhuvvB2NgaQuqpijMwwTAo1fZfpHTt2jo1NvPnmgV7p88xVVY2mSnl6XXsl3EbDJoGvEgZRkG+88cbOnTsTdZteeZNBRbI8ZyIRIUBCTCteEXHO3HnnnTHGAwcOVHUs8mUJ+jfn4NXYJPBlh8To0yft+yApQKBer1pcXCSiW2e2QpAZQ7SBYpWvGlIWL+q7Fltrkwu0dRxCKHL7wAMPnDp16viJEyGirn0q3lhVNRHqOmzOw8PYJPDlRwocFZFh30ARFEX+0ksv7dmzh4iYEeON4GF2uaAiBCqc7fZKAD/+4R957bXXzpw543IHYGGhUxSZ9zHLLDan4iFsEvgyY7iyceOoqJQc+k+cOJll2fT0tHNGFeamLNe3ll03uakxgBBktFUYQ6J4/yOP7N+//63jpwQYGRtZ7NUp+S4AEd2chxM2CXyZMfDcbjyfkZJsQkT379//wAMPJG1qCPEmFJ5XYqjeV1oPO8sxqiWUZb1lanznzp379++v6uiDtluZCLxP+a5v+kfXxyaBLzP6PkY+uTGGEFJZlCNvHStG2q1WwUwxSOaWCizeVFg2cy4VdkkVrgWAaAAw2sqqnt+9a8fu3buffvppIvJBwWBnyrJeZ+i74eoqXQQ2CXyZkSIf0zws2kQglVV96NChe++9t9criWAtp5K517itGwnELDGGEDLrqrIG0Gq5GLFjx7Y9e/Y888wzSKHNQFFkN9+4ty42+9AVwSCKxRijwPHjx0dHx0farSzLkohd1+UVTSi3kSEr5uE+VNUaG0IoioyAGBUAE3bu2jU5Ofnaa6+VZd2cQTYZ3GCTwJcZMUZAUoUE1aiqZdk7evTo3r17qzpayyGElP85xLqfBOOScEmxdQyCUhPMN1weL1VwWAG6KoF7K/ucMoAYgrFWRVICzbqu2cAYhCjEevfdd5dleezYMQVKL8asNfAp33wFxDdeSp0rjnUqTeuqXtsvTL12JqqUAiZ9UFWmpFYVJVGwQoFomGpfHzt2bOvWrXlu0/E2GUIgbEkRCbiUYXQFu5YOXC+1jQIEo8OzvSqJSj9TTZO7uTmtDJ1tRU0WrCT2RaX7WbtVqzaxsRmGEmClNFoKYSMKOGf3PnD/d5//nnXZzp23BQUToigrrCGRpq2qWDufZ3PVFRM345Kbv+Fw041YlwIGGnXnWu8CSKqpmbLAEqAgwyZISHNet1c6506ePLlz5841z39Zs0CuiQHlZHCZdNUkxPLwbhclll5NRZEAYLCICCAR73//owcPHj53bp4IVR2ZyRiKAmaAoEEuyN4bEzcfgdfJcUHgFa+hrxJXBf2JiCAEowKJEdqU4NaYGG0NF4BVte3W6Gs/2D89OT0+0mZFkvD6500fL1UC4uWvlbe1RmYMEkVMtUpT+9YiIQ99aIat5qXrJwW5qGwh7wZNYyQYRIy281DrBx//wIvfe+n0qbk8a8TopApUKLm+RECy7DW0LqBlP/SNQO2bj8AXiYGapXlfka6VkSoPGDMoHZZyPok0iaa81xBw7NiJu+66J0YFKSXHyj55KC1PLxkXH92+xlSpK4w4S18wMLyGbGhxwWZccSTRPs8MAyFgZIRV9eGHf+i1116dm+8A8MGngSnEEiSqof+E4/LXIOHmjdbhb7T7uUQIIA1XV7wShtmbdCSDV5qGB4qT/lFE8BFBQIZeefn1O27fk+eGiEhBEEKk5sjLF8mwrNnDs+tQs5doT6RMutwTQofZvLyXN2POlUtf+TYwDI1qDZxB9LFd0OTUyO49u1555YWFzpyzRiQA4qwVDWAd3CaAlU9DhzTgK5/V9YqbnMBD0HUKfQ0krea9EcxEg5AHBVAACVhAMYRaSYxBFK3rMDs7e/vtO/vEHlBkWQm/S2/nJfkQNrxVJRVavnH5PS7DIC3z6j2udnYbVbVGRYRJnKMQa8N667bpLdOTr732Wq/sGXZ18AoSkb4EscJPc8UZMfTVdc/hm5vAaxgeFKrDpFVN4UQqiEpeEBVBENgIURRUVehE7SoqUA1blnIuYqEKc0ePH9y1Z0c6/arEuPJuJ7TVwsK6d0QALd+YplQoNEIjVKACVUQlDwQgAuGK9WxZ9boQVAIg3ldpZ2NIEETkvnvuY7JHjp3slD6qE1hjCoVBk++aB/HDzas/Piv6/4suS3t7feLmMyNdACRoxMk0aUIBJRFI39AiA5kyInqpu91uWXWzLBsdHWXi2pdn588w21Y+efStH/zIhz5Cqr0ytlrpOXP/DO+mkUC/lWufackm1OiblQCoLN1XTVwPbiTtzZD+epz6li0L0iHmE5TfkeH6XYHZdTsL7ZERRfTRs7EAcud89Pfv2/vyi68Eifffc48AosLJ/g4AzUgkEEYEjJIS7JLSgd71GLoxcNMRWJoiWrwilkARCSoQJtShtNYIYpp40zTcrRaPHD1w9Ohh68yOHdt23b79K1/58tGjh1NVrttv3zk2Nvbq6696qQi26vEHHvsJol4IoWiNoImAIyJu+ELyzrwOGjP2WrJ3XVVZ4aqqynOngA+VsRYk6aZK38ucFURb+Cqes7Y11z3vfTUxMcaAAXfrbqfTabXGJotJwFShW9hRH8WyI7L9EL4l2+86BvV3gJT4aqmq6IoU6nk2BiCGwGwAeF9nrmBDgO578L5vPftMUdgdu3YwqUAW5meNRdGyGRmPXqdchBhnWyPZlk5vfrQ9QbDRB+MYIBHh67vk1s1H4ISVkUAkUAFpVS1kubU2LtSz3lcnTh3L2m5son3ixInnnn9mdvbMxOSIMfzi61+L0edFNjY1kmV2sTP/yv5vGWPGxsYchVZrLOvhq09/YceOHbum7w3osThDDo3auT+PvVMOr4lOZ35ktB3FZzl76TFzQMWwQChjtyy7J06fmD1zZnSifXr2zVK2/9VffjFE3+0uZjm/733vO3XqxKFDh0Rg0dq1Y/cHHv+xydZMFRdyOwJo8JU12ds34jJhwF5VBSiIJyFjKaACkDktcT7GeObsubOnz8B2vvz0f/yHP/uxUyePP/udb3YWZouW3bFj+9ZbpmZnz7xxYH/wcsv09rv27HvkPR+M6MUguWtL9Gwcm+teL003W0CMIBKIVphJSEKsjImCutLFk6eOPv2tr5w4fdxkAhYfugBGRvORkZZoDKE2Vufnz09tGSfSGH2Wu7qu67oeGRuLMTpXzJ2rW3ZL55x+/GO/eNvMPSqZpQIY8gBckkUvrQPpkjl6yLjVWDujwHd9p+XcbOdMp+yI1tMzo4ePHPjKV54K4qEKlrrqZYUZGW0DwVgC9NChQ9PTU5OTk0RWanv+XG/bzM6f+dg/apvpKAyxmR2BJhGagCsyA6MfyLWsEBGJAgTp1fPder5Tzi10z09tmZgYHf/Wc9/67veed8ZmLVv1FqvQGx9rKcLYWCuGamFhodNbbLfbU1NTgJk/14ve/dBDj/3we380x4SoMSig7mrEkyiScmGuPvzZ/+tf7djdViahoOoBWClIWqffCv/0l/67La07DY1i+UN4W9x0M/B68QPWcBU7MNWbB7//xSf/YmK6PTHDRTvv+UVjiqruVdV58p32SJEXGmOY3JpFWQwhRPG+4jzPi4wWeqezLO91FkcntviyA6dP/O2fPfrwhx/c+5ghQwqgL7FdrvVkc5IoqOtQwvpDx179+jeeqkIvwmctUtS93uL42MhoK2sV2WJ3saokSmVyVUarsJ2yt333WJ4bdr2zZ89PtG9pjeHIidde/sFzD+x9f5tnmE1Vl7m74jVyiZamE0VMi3lBeb576nsvPPvm/ldr6Zg8AtItF8uynJqZnhjL5zvnxkdodv4c59XoWFFWJ2xhxwsUdabqI81ZW4zPcPD4/uvf3r59655t+yyNEeBjpEAuu769OW46Aq+BpqMEUBBUX/vGk9vvmPK6mLV4sXMyK5wiZEbb4y0iqqrFGH1eOAWqulcUxVgx0u0uVr50zmVZpkQw2qnm2/mYzaNB/bVv/M22W7bfNnO3ElNSk+IyOQGRqCooAkFROxu+/fdPf/OZL8/smGxb7lS1yxCj7/g5k3MlHY6O89jKIcLgkg3KoEJ1e6zo9Ra90NZtE2VnsT01Qjb/8tNfPHL42A8/8pO7tt6bZXmT12tZs99JOdK3uaGlzJPp3Xfl5Be+9P+cOXt8Ztt4pnXQDqBFVpmWd+3Fbpgb21LUVXdyK0lYKGWxNd7qLC7EKHmRG3Z1XXlfOtuqxHcq/+RTf3H3HW/+yA9/rG2ds2OXseXXCtf3AuByIRXKJcaL338+oNetz0XqRp5X0/FxUeGj9Hq9hbruGkNZbpk5xphleYyyuNhJxShFCOxcXrA1ZAD2pghj43Zkkl/8/rN9vejSJd+9QTUJmQPHo/3HXnvmua9tubV1vnOEi87IRFAzb/Lqlu2TsPXoRFaGhSoswNTsItlQVotV6I5OjC50F7JWZjKz0J0TV3aq0/mITM2MnDl/4itf/9JceU4Rr7T+WSQ5eyY7niRBKaL6+jf+dr53fGLGwC56PWdbwbaqfCSOTlq4Srmamz/jpcus7TEHjvOLc8Zx0W4pOEjk3BjHVVgcGXe37pzMWmH/4Vde/sELEVUVL28VyGuDm47Aqz1gk9QmiAwsdOdhQt425CrRjmvBOjCzMc653NoM4BjI11qVMXOjmR2XkGksGCMai+DN3FzZao/HGBd7i9Zpr5p1eZhfOAnUQz597779K8gvgH/l+9+1hbbHeXImP794zMvZyIvCVbdeMBnK0AXHscmxxd6ikNQh5K3WyNjYmdlzWdEqaw9ySojSI1PXsjgx3S5G7Fsnj9Z1FdVfzJO8WKxt/WpIq9p8CKFe7MwdPXGgPcbs6iouZm2KKHt1lyxVwRubsW21R6fYjAryc+drUCsvJqNmZcU+WkEu6mBs3m4tdBc65dz4RMu6+MKLzwo8842g/bnpCLwCmuJ24WcXj//FF//dd/7+a2S9aBWjr6OIIISgGgFRjf0kOExkRkcmy15dVbW1DsoSYdlZk+UuX1zoZrlttVwIpbLv/f/svUmQXNd1NnjOHd6Uc9Y8oUAAxEhQJE1S1ERJlGVbcui32h3hrVdeeeuFV15r4WUvvHF0dDs6uqPbC7s74o/+5ZbUtmVblihSFCkQIMYq1DzmnG+4955enMxXWYWBAASCINiHGWBWVeZ79953zz3nfGeKW8VS6O6ucD4sPx9NQiAggsxC0qfGytaN+mTUSxqWemEkLKZKE4EplQpxFlubKd9rNptRVLSW0CGA6HaTwI8QpFZhHKdSaqU1CUJFnX4DZKp8t7mzYo8m4v0WdJcLCQAQOLDmiIDAEaSpa7TT7WZvB5Rx0oA0qUmstX4YpqkJw7DfT4io30uIAEEXw7JzwiROoPI8j7Uka21mXJxkvq+DwGt2dyv1Qi9pxqYjkBKTfdaZ+HNnAztrkd2xJLIMhAChTDvZ/q//9L/0zMbUrM5cPzNGeSFKkaaxEo4BFUQgssA9QQhsZjiLEIhQWADH/l0lvdRkUop+r1sKA5e4zc1ND8b7Jikq5SwIBCAw1iqPVV/voY5RZ1MhJYDo9eIwCjLKLHabvfX/9k//RyvekHHRD11mTej7aMlkBlEkSaaUBwBk0FchGCFBIQJloFE6DshC0Non54QK4zitlJC0QeglWadc8QkyDgg5LHIfyY99AGOPeNFwsJCIUgjhIEmg8dHyz//9nX+yqts3zvOFMQ4FKuWRIYXKZU5Lgc55SnAUGYETKAAsOAOYAZKQQghBThCStURktFa9fsvzQ0/JBJJBFzX6DFfn/txJ4EH+EBEAKAVCgXHp1u56p7stvcxBPyhq7asks0lqpfIP5ehgnvXuDmxCdABOgAM0SM5aiygFoK9ku91O0/TU6ed7SW99fd0CDe6MBzV3HmH8ABD3kjAKCIjQdbr7/+v//j850QtLwg9AagSALHXOSgQ9DP86oDyJgv8VBAAOaRCV1ekm9fpElmXGxPuNrdS0l2/f9PCTCXfA4TKSyNIUURrjMrIE9t1f/ee//uyHfmTL9VD7AgQIJaWUzjnnuNKYE4MrWE4UAQAkx/2WAQDJOmestUREDqWU/8/rYwAAIABJREFUQgilRaOxu729vrG7KgDlJzSvJ0ifOwYGAGB3xbBLvbU2jmNjXBgWhu8z7Qml0ZEFUgAKiF8CAAj55Ug4Eo4QCIUD5UARKkShpE4SSyTG6pO+HzpnlBKN5jZBisIZC4CAAmyGDyt+B0SDghVpmkrAX//6HcAUMIsKGoW11hJiZpHAB+EDwEjUsQEYDls4Em6YjMGvFNCUCsWdnR2+fqFQmJ+ff+edt9txG2CYEDmgx4RpDQ9H7XkAIKWQiImJP7p6BZGUhiBUBMZkfaAM0RJZQwYVjUyKn4IjNISOc02IEEghaIGeEEpKjY7iOAZwpXKhPla79tEVIuvoM99A/HPHwOQcAKAQgMg5Bkp59dq4ECqJM88PtdZCgjGptYlzKQBnCKh7VCQfXhaBUAAJpVRjf7/X7knw+u0kjQ1ZJUEWCgUFAGBpiEUjykeJxCIkQ0KizUBKRIDVteW5+UmpbBAq61JjYyGUEEoI7awYHEAHxGxsAAYcC5jn0AIAxHFcKVXTOJaowrAgUfR6PXH3Qsy/TcLDQPAeupxJCUkAxHFvb29ncmqCrENHR9qLjqi7IwcTGkDLfjVySE4iaCE0onCOsixDxCDwEFFKWQijnZ0dD1XeBOezS587BkY4YBtEsBYkyrHaxIXzL+ztNRqNVr/fV0oRWIJMe3igKg+zat2h13AToeOXSZJCEDSbzSS2ni5JLAooOKtnZriqjpNy0FVA3LUy28eQABCoPJORUqCkspCdOHH89sqtNOsTpIApgBMCEIWxaJ0E0kAaaCjq0QFawAwwGya7D4KckRSS2N7cUUqZDJwV5IQUfhgWuFbucB3gSF7Hoz6KQ+ScU0o5ZxzYMPInJsb293eJKEtJCU+rEEEDKCE8RGktORRukFrtAC2ABTQAFqUkIRwiy1ZrKI7jdrttjCFCa0mrwBoX6MCC89STiw/9hOhzx8Aci+Cs5bdcTkOBHh+ftgasgV43sYaUUp5W5CyAOSjOcte6h6N1W9B4vkBws1Pz2+uN1p7x5dj+Vv+NV9+qBVMOBICQUhI5RBhWkHhI4lC7AZwkAPDiixcKYdRo7LVaDecsoGO03DknhR7a8Hions9RGv6JRKlQuH7lZqALWgQ2Vd2Oee746Swb4dJPBu4RQgA4KSSR9YX/5S99qdvqdjq9pJ9aQwiKCJ0VCBJRDidyl+RkGhyPxI3jOD1JEOzvN20KkgKbirTvTp062+12BXzmu0N+/hj4cHq9EMBC9eaN28VCbXrqWK06BSStIWucMSmgBUwH6uKAV0c2DR7YloBWgCWXlKIQLAS63OuI7Y3kra//8YvnvgoQAQUcSolIw2S3h/UMO0ACcFKDc0BAEmQky1OT00KoNDUAQOQcGSGQKJMDW3EwVwfCgQKSACMvkoJg8AKolMqh8rvNFEwUdzDtyi9/8a2CXxms3aHCHfhbxKLweXfoOEizQdnnlNJadSz0wzTO0BFYB9ahI0GAREBioEZRjk3owQskEQE4FBbQAGZCUOirKIqSnvVlEUzU2stqldnTJ8/XClX3/+cDf+bIGKu0RCHIOhQCCQiFBKVVFPmVbqsDEnToaySBGaph6vtB+qgYgaAPlWLlcH9nXGJNIag3bPPNN//gxMIL5WCWyHPgCRQwyJVDAHLkHiWYHglMBtJHZN+WSsjeuLo8eWy2m2wpqQxlkkh7mGXWUXJUxAxqX+U/u9E3SGAzmJ5cWFnetb2kHM787ne+UwjrEjwYFJa+d1mPh6C7sw2vhkSN4F25vSqFP1WbSqEn9CC+EtESEThHB1lEyGfQUJ8HREBBAOScsRYkSETlKa/gl3c2u5S554+/+OaXf19iBKDu9Ix95uhzx8Bq2OEO5WhQgXzhwss3/++PUKtStdDrNnUQkCNyBoT1PM8YkyRpFEW9Xi+KImMN86GU0lrIUoqiYpY6sqBREsi0pyqlmReffwMgQgglaAGDLHzWFWHQoOsR2MCRAHDgCCSCQ2GN9HU56xsti56QJu06sL1ex/NVZnoSCBG19vu9FEB5OnAWAcBaq5SSAtKsH4Y6TnpEzldBaoXnFyMPvvbGH5w++aqkoi8LABoPji0cHAEjCcIPTiP8ctgJDJahMvZtlaJyu9Gp1KoWOuRSFIRgyREI8AMvS23mrEDP87w0MURCa93vd/1AK4H9uB2EKjXW932bWZOZUFW1kBOTz1288KXjc+e0KCnwiVDi0bTwzxx9/lToURo+OwFqamz+m1/7vWIwfvvmTr8NaQ+zvvRViayfJkBOF4vVXi+uVqtx0kvTxNfaOZclmUDPpnJztaWxpqjiqfFuS/W78kuvvmWdVFCgu/QhFIOSFw/PvQAA6FCAlGAtWAORrlw8/5rpK8oK/Q5KF2kRhl6YxrGvpB9oY9Mk6Xu+DEOfwFqXCSGCIMiyzDkIg+LOTiPUhVAX0z6VC+Mbt/dffuHLL5z+YiBrvqoK8N0hif1YrMajE0dAMazvKUDXq9OLx85sbTaclZ6KTEJoZTGMbGY7rZaQEPkeoInjPsONzhlEVEqlaSaFn/atFr7LQJKvMUj7kHTx1Ze+enbhd3xR9aAoIBSoPuvcC59DCZwX6QcAAAEISECIHoSnj71w/NiJVmfv5+/8542lD2tjtaTTDkvVbrclpWx1e0p5nVZfgIyiIO3HKIQWoUl1ICt7rd2NuO+raKvfrVcmLr7w6oUTrxH4AgQAknMoRgMqRpTYh99CiAN/rLUOlZQQvHD+tV6v89GNXwtPp7ZVG4tM1i+GhTjtJ30bRSUiSJI4dWkYhkqIJOkQSQArQe3v7FcKk2kPpJShrm2v9Urh5MXzr3tQyozHIAEOihC4gen7+DuYDM4Fa60QAlFXimNffPVrv3zPrWxd39/ter6qjZf63Z5WYblUbfdaFo2WymEmhXbGSkmIcbeTaBV4utBuNhUEzjlED8lLunTmxEsL02cQApMK5SmTkhQoPvP5/J9DBqa78AyCilPrexGlplac+9abfxj+vHztxnsWZK/bKddLzWaz04n9QI7XatYlLpFhEJnEZQlkqep1zB//lz/VWOp1s1qlUq+OFXQ1I5KEqc18HQDASCqSuM9IPo4EgAUAIAJEIYSSwjpTK07+wTf+u9df+8rK+o3/96f/RCkKT6LNJJHveSZ1vX4nDAMEkySJlFJrlaauWhmLOybti439duAVlNQuzaYmjr3y4pfGC3P9BALPdxZQ5lmEOXB1sHCPREeDw5gcOSE8RHTOEXjHpp+PoqCXdW4sXXn/g3dau65QLlvX72Z9zw/TNM5cBgBCCIbcPc8TpNJYZinubva1okqp2k0sEBybPfX1r/6BgDBNSKsQCYRAIYeJLJ9l+vwx8AHRQTguQeCVjE0kFDRoDwpffe13z506e2Pl0o3b71++9H4UFePY1cszrb1USl8WCmlG/U4a+CXTN9/99vcWZ896UAIQqUlD5WfWKuEpoUCAyTKlc4P7YMuPRAU/FEn+LiIIJAA0xniel2amVpivnZodq8//6Cf/Z3NvMyig0EHSt2ma7e93Jqd8z/OEAI2aHARKJV1o7GTV0vFv/5c/XL+9o2ShWCxPj01UojGEMPJD4FZDCM6ZgSQe3TDcA+HxiDDnyFhLWiEQCBSZ0VKE4+XjBuz02NnIm3r/0tv7WzvleiFJTavdATS9XrdQKJQroRYqTVIAUFKHXmVjZf+Lr/z+c8fPtPa6zsH83EIhLBd1zToR+hEQ2Ayk4rs68RkvivV5ZuBDlCVOSB14IYBL0iTwqnN1f7xe62f7jdZupTyOoNO+++a3vnH5w4/WVlfTNK2VxyrR+He/9bXx0rwFSeCTgVAFAKAlAIgsy7TWSuFdqkg+quwyxiiVG28OAKVUCMrXVedcnCaz9TPf/8M/vbH83vVb7129+UEv7kbFME2EM37qABElFNLEKhXsbnfQld/83e/NVc7MVLQl4aEvwRHILEOtB73CjEmlgoHdPiqxkJuY60fiYXEEABcouFpJloFWoJW2xkgVKlCpTV48+6Xnjj//myu/uH770sbqUqHsx1lijJgYG096hDoQFAmhXAKbu63nj7/02he+WfDHcTIkQoVCgLM54oyDbizWOCk/4wr055CBh3KPhk6RwabU2gMAZ8Ba8L0CgEtNLJXf63U2NtYau71ycers6d9ZmL54bPqVZrP963d/9bvf+FZmrZRSQmhSozwfFQA45lsi1NLj+9nMcIHI4ShG1emH20OIGgCJABGEBADuxymyBIRUkafSjAre5IXTX3zh9IX/8X/7H2bmAMC1292L51/u9czNm8utPSHAJ1G4ePorczOL2+vd+YkInfCUL0BkmdVKawXO8EZ3QuYtFEcKiQkHwLVp5eMQwgTgOKOQhjCZlB6QywxoHVmCSnH8q7/z1uzi7M4/7NRrFYfOZO7MqS/cXl7rtvrNRsfXQRiGb37p62dOXSyoucz4LtNBAESQZcbzAJDIGRQKpQACKRHEo5gxTxV97hh4SCO2HD9BCyBBSBBCAIEjITBwYDZWWr4c6zbd2ROnXnv5TQ1lBYEolD2xlKZe5AVJ2gdPK6VySFtrbYxTSvFetNYqrQC4DhaMSB46wNIemLg1rrVOKS4xRwBAzmlPAUDcBy9AFL6FQi/rt/Zdd3O7UPQR9GTtubnzJ994RVkjW81+rTKhZKCk/5P/5ycCQq0UADgHioWS49MBTJYoLRy5g2DoQxv+t9n9o7NGR4NOkJ4P4MAakFoCotbCOSeFT2BSSCnzex1Ik1a1XhOoXjz75VfPBt04+/DStWKheu7MaQmJhqDfs2GghQ/OAjnwtMqyRGvJiouzloikFlmaai/4Labw6dNTx8DWWk61S9N00CSWiIENRETE/APOORgpQQoAWZaxenmksPCdxF1/EUZqu8oBMkRIACgQyCqAwvzEy7eud16+8NK3vvp71XBMgWcdhCqo1yYb+x055qP0AYQUzjkzvKlQSgAMut2qo5mDw0jswz8+FPH1Dy4lBuumfUtICJBZ8nV1ezNWXkApvv76F+enzkdQNkJIL6hNqNQYTwZplpbL1V4v9n3f03qQiUcHtfeUDgGcwMGKjTIsgs8x3Y8w/sPECf1+fl0QIMWBv8DYWAlFoMh5zd047ggUpF32/e9/z6ciOTUWTLx4pv7B+1cUVT10BORrArIopBBgwQEKrXU+fqEG0/uscy88hQwspTTGMKgIXCqJa34j8htmkjiOg2Cw+mmaSimllFrrI6VJ70287Yb1mQc0yOweXEMqAeLCmdeqlZnTp56vl8cliDTLlJTGmlq5trOzNz01aR0BgCMnkYaOlo+lx2h6DYxJKQnAWMoUSgcohW53kvH6fLu99aXXv/SVL78ZiSqCEhZQKAARqIAItPaeWzz+4YcfvvrqqwDgWA4ehabuOloxBOUey1wOX2QkYJOcQUQBQoIvhAiDaqBKhWL4+9/+veMLpyVpJf3YoITQ98qt/bheLQBZrRnKtiiQSzIgfLbBqnvRU8fARKSUYulKRNbawdkJkCQJl0oxxgRBwAVTPM9jVmcRlHP7g91thI3vGAZfpNlsPXf85Hht0jgjhOAThAjGxsau3bwBcFYIJALnPmVExBinlNBKA0BiUillsVj65je/Hfdb586dKepaapySQklFgAhIBMZYqeX01OQv333b2kwoDxGfNqMQkZV64cAAyNmZhVde/uL5C2cW549JkA4IQCGC73v1en1lZWWsdsY5AuC2r0MLhejpmtXjo6cOhXMjjcAQUWvtnLPWdrvdIAhYhebsNiml53n55619DMnZNKiRNUhAzTK7u7tfLlcIACwZY/SwRkehEBBRnOTJvYOqML/9GB6NhBAAwhEQkFa+RGkNJbG5cP6lamnKgUJQAj0AkaWWR8qZOgQ0Pz+/sbHx1LpEESUREiGRiMJKrTbx3PxpIpVaQtCZtVr6WuHExESj0UjTTClluXASIruIn4GIq3vRU8fAuRTlH5l7pZSFQoGfCv8yjmMASJKENWrnHBvGXMrM3dEO8GOJWZYIAYQDIgQiaLfbxUI58osIUmtfKx+GJwUijNXHd3f2+FY8jE+pzYUAEEIMSg4wMsCD2tttBH4JwBfgKxnyULVWWWaFACUVAFiXnTj+3PKtJa4+c697PHoNyt+OyPHhiAianEwTZ1LMHEj0PRlJ9MgJAeAIokjXapW1tQ0AyLJBpv6RYgDPHj11DAwAOVjF3NtqtX784x9fvnz50qVL//Ef/yGlzPEtpRRzMiJmWZYkCQBw9aOHvSmfGqMP21q3v98sV2uZtQjIfJtlFlEKAUQwOTm5u7tLg06Gctie69NZUh65QAEguCqFVl6pVPN16BxaC84B6yiIoLUEAEfGgVFCBUGQJFncT+VdC298qsSFA51l5UaYDAqFihYhggcgjXWeDohAApCFhYWF9fV1a4m3hzFuaPI8szz8NDLwUMShEEJr3el0fvCDH1y+fDlJkn/8x39cW1vLK0I553ILOQgC5ltrrTEPnSjPsperPiFIgSJz1Gi0ZqbnpNQIKKUmOtjfiFAoFNrdzkE3EHbOfkqEiMYYY5wUUqCQQm5v7zkLjgAdSAlSgtZgzCCr15EBcIIc74DZ2dmtre3hxe65Kw7L4ScRSWzNIElQCpAC4jhBUGnmjAVwguyggB1nEZbKBSlxe2cHhbDkCIEDQR1Qnj31jNFTB2JBLkyEiOPY9/3FxcUgCE6dOkVEf/Znf+b7/j/8wz8YY0ql0muvvfbDH/5wZ2enUql897vfnZiYAABGth7t1geqO4BzrtfrlYohAqSp8TyVponv+0TgHAgBQogsyxyRBZQDHfxTgzqJSAjBAstYcM51u93FxUWFAAqcA+dAKVAKgMhaQ+ikkABgrNHKW1hYfPdX78/PzkXB04XWsi5Fg+QNyDIbBJGnBwGQcsTRqAUmJhsfH9/e3h4fH3eOtFZ5T+9nlZ46CcyhggDgnAuCgN1ISqmf/OQnf/3Xfx2G4Q9/+MOVlZU//uM//ru/+7u//du/7Xa709PT3W6XufcBEYs7JYnI024AQECS2v39/Uq1CgQE4HkKAHzfBw6BEgAAWuuZmZm1tbUkyRCQrfdPYEkelEbLjAgU16/fmJubIwICEgKUGswakaREJSQCOeJALgjDkAU4t7G35PIKBncCCjjy+sQnJcjaTCogACnBGCOlJK7/Tmzi2kFhaLCeEnPHFrZ29ww5h2CdBQBjDJdSeCbpqZuYUooVYJbAUkq2b996662/+qu/QsQPPvjgxIkTiDg5Ocn8s7Oz8+1vfzuOY05GO1SB7YEJEY2zzg0i+ZSSm1tb4+PjBx+Ao0e5EKJYLO7vN4JAWwCATzk7fIjwAeJgrwuB3PrsSGVZABi0QOD2lwAIMDMzs7q6SgTWWYEHduOnakM6QMKRHih3D9FBh8PyekqpsbGxlZUVpaQUkoi08p+B0jn3oqeOgQEgyzLeMWzfrq+vJ0myurp64sSJ6enpb37zm5cuXVpeXo7jeHFxcXl5WQixtrbGcR3GmEcwgIHroA3Rb2sBEJrN1vj4xFEpM4z6IAKpoFqtNptNBDCGoe9PZT0dgKNhiWNeOmNJSm4d4QCI6zeORnE6clz1LS8qc2x+YXV1FRHcAOsamI4AzrlDpWc/FRoEvNqMg0m5ICEiIRLmPwEiwtzc3OrqOgIYbtgA8PBOic8MPXUMTERhGOaiLE1TIcSf//mfj4+PMzr1ta997c0331xfX//Lv/zL3d3dN954Y2Fh4W/+5m8++ugjVrZzWOvh7gskpQQBjsARtVqdKIq0Hgxj+L+RQtAEiBAEWgjR66eIkj417gUYiF+XV8Da29sbGxsb+dgRETrUpvPzCaFQUIjU7cZSygF7D7qN0ZNohH1vQsTB8YSDSkA4GPLhjwEAgLWuXisDQKPZISJESUBsJjyT9NQxMFtcuftXaz0+Pv5Hf/RHr7zyCgBIKX3ff+ONN1599dXZ2dnd3d1KpTIxMfGtb32rVqux8pym6W8zAESQEtfW1iempw7Hhoz0VQGAIQ+PjY1tbGwoCYiPsQXYw9IgcA0BhQTnaH19dW5udkT1xWERH36JQ6GF5AQRACwcm79164YQAzE+0DyH1Xc/LcpPGSIwJtPqztjyg5cQggBmZmdv3rolBSJCZu2zK4CfPgbmeAwi4jdDndaydBVC9Pt9GIZh/emf/unrr78+Ozv7J3/yJ9Vqla/waBA0AjrnjHUcC723t1er1fhP4iBt6OhOcA6mpqa2t7cBYNhx6ROmu6Cq7GgZOLF5u7ME5l6KQ+49Et4sAMQwwtChIOtoYWFubW2NfVEAkBsjn54NORzn0IGXY5xHT5Rh1WuBkGY0Nze3t7fHTxMR6dkFop86BmbKE5L4PUNZAOCcC8MQAIwxvV7P87zjx49PTU2Vy2WtNUdHw6NGROEwtbXT6SmltPLv04GMiDjqoVQqxXGcZu7TDhZwg5g0cgCM1g7iKwfV82jI/G6QtUGEApAcN/VzBJnv6VKptLOzA5yecf8ObJ+GgybPRTs8kqPb2Pfk2NjY6uoqHBxtfGA9a6z8NDIw5zDAYVmaO4c5Sk4pFUVRHMfsV+CwrfzRPgIDW0sgQEmwDprNRrlcFuAwT7o5THmNVYGglQCALMvME9ka3FptSHk6sSBCAkGAeZalyczdn+4h4XWAMCsUxpr5xWPbe7uJsUQkhSQgR/Cpu2FyCZyj0EcXO5+UA18jESweX7i1vMQusXz8COIxpU89LfTUTYb7UI2esrwjc1iLMSr+QBAErFfn4ZN5RPTH34lgxLkyoDRzQsDOzs74RM3TAiwhgrhrNXNHAriZFkxPT29ubkoBqXVEAO6TCh+g4XDvuLYgks4Ja1BItb6+OTEx4WnFgDLm1bgOO3B5SaXUQChACpBayqnpmZu3lgkECpkYCwAChclyMOCgG9QTC48mIi6inaZGSuWGTdkPjukhuIgAAgEsSYAoioRW+522Q2FhmI9EB4rD/R9Rng+X3+URAuyfAD11DMz8aYxhSZtDoPTYCAY8BhYccZQtkZUSHYHWIk5Mq92olisSSUq8140ZCxUCPQmVSmVvby+z4Om84uShvfF4x04AHMLAiDk4JCKB4ByghCRxy8vLCwsLWZJJIZl76Y7x3P29QyKYXVjY2dnJLGgljQPieNVh5MRoCMUTIBjowAAAga8ajYbneQNHNo6usxs4AgHAWUTwlJqcnFxaXnZugGsMTj4aPI8HeTQwEpn76ULx96KnDl6P45hDoHMGBgBjzKM5h+5NObLKvkOwDow1Qqper6eE1Fo7Z0CAHGhfd4AmRERkHSglSqVSv983xiopB96ZIzd7fCjuoUEPf+LSOkTEtdHjOK7Xqmmc3uvWo78cfS8QyNDU1NTyrVv1el1LKQWSHdRUuGMkT0L+OmuElNZaFBIIhBBsRmmtRx7iyFwIMmt8rQBgamrqnXfecZlxIKQYKeiFeWntez4atss4LogNNA5oedoyE586BvZ9HxHTNL169Sq3Y4+i6Lf0DI0Se2sFOUAnyDkEAEEgrAMSmBnXbreTOP7ggw98LY0xQwa+U30SiGgdWGuF9ra3tz/88EOtpAQSn1jHO0IAEg5BkgMABCcIHAhCAFKZNVJ71trt7e33P/iNFtLY9KG6hxAIkqqfpteuXSsUCuSM53kaIY5jdUcz1Cfj92bfHKLOyGWpa7Va165dQ0GKEAUJcjDkSkIBAMjlJ6VIjdF+sL+//95771VKZXQkuF4BAIDjjFE8+M1RUkr1+/16vT4/Py+lZMvuaeNeeAoZGACMMfv7+ysrK91uN0mSMAy5HPljufgoA+cgEDMwKp1m2fLy8skTx/f29nwtnXOHH/BosQEJAIAyTVPlB7u7uxubmxPj44IcwieVQX4nAwMA0ICBLTmhdKfTabYazpliGMVJjw+gBxkPEREIVBqF6Ha7P/3pT8fqVWNMpRDFcSxGS2KNoL70CW9pgZQzcJrY1fU1YzJHRlgCdHIQDcojGTCw9r3MmtQY5flJkvzbv/3bsfkFhYIZmMNCcizwXgzMmKgQYn5+HoaW3cfWWnvy9NQxMHe4mZqaeuutt7gchxDiMS7cENDJ/boD74K1kDmXpOkHH3zw+mu/IwSaNNFay0MA7GgkFlprpVLOgSHo9XpXr1178eIFT0qET0rRouEghofZgIEBwTpwCALh6tVbUSGcnZ2SAATEG/QBx0MAqXWAIsvMz372s29+/atxkka+56yVoxJ41G3zCTMwOUMARJIkZin97Of/+bWvvmGc81AA0mihaj6anbVSSksOAJLMak+/8+67x+YXpsbH89Zpw3ZqHzN8YwxL3SzLRoHSp4qeugEREbtz2V/PhscntnAHl2UB32g0C4UCYyFCiDvUxgMUOofKhQAhQGvZ6QxzgxFzsPMJex2thbWNdXaVu2Fg5UOdJkoKAqckpkk/SbPA9wAA5aFopydZngPZ6hWIw/g8AJCD1LG7DIJzwgQKgUIpBQS1SnVra4shv9GRf+wklFIcgMBv4DGVbXq89DQy8GjkRg4GPsJ17gEsDhDUQ2gqkANAxK2trcnJSSUFAmmpOJnxrsR5i9aStYQInucRUZwmwHDSYc69P9T54MRqLh6M+mAKiCAQLEGSJPVaOcuMFIfufmQ891k6KYRWcnp6en193RE4uvsiPBksGgYpouAIut1urVYzQybiR3loXkBExKCxdVYLdM7Nzs5ub28755wjNxz1gzya/K/5VnxcdtxjpKeOgY8I27yO7MNSLnZwSIP3w2dHo3+GgcxsNBrVatXYwemOI18/QqwaSIkoCAGkwPn5+dXVVecGyTF5+GFqsnte5VFokHSTvwOBiGjMYK/zTXOJMTr9OxfnTuKQJWPdiRMnbt++ba21hgSKzDg6PI1Di/uJEYz4Efk8ZaQ/n8mhefETJkIAJaS1aGaPAAAgAElEQVTJrCeFQjE7Nb20ctuQI4TMWFaRmM/vd+t8d3zcon2K9NQx8CdH+SMZ/RUMoyMAoNFolMtlKaUc6Fcf47hHREdO4CBAr1wut9ttlAPWFSj4uH+0/OSHIgJQCq2DVqvFMgrdIHDtoa6DABIgzYySolgsJEmSJInWMkkybmuQc9Hgvk8wtgEfoF4/06icRADf0/V6fWdnBxGTNJNacqJ/Xr34M03PMgOPHp93/h6PHt6wubk1MTHBeUXWWrjH14fFHw+lFjpHtXql1+sRkHHODYvyEBF3IP6k7WFEMMZub2/Pzs7SUGqJRzJVtURyVgIcXzi2unzbEAitsswa41gjGmiYiPhkQZ08EPo+K0kAYpgPw9NHhFqtlmVZo9Xk9i25KfuJ+fueHD3LDPxQZBzs7+/X63UEEOIgku5B2I7ZXAkRBEGj0Rj5PXIULj2SGf9QxDb89vb22NiYZtcvQW7DP/h1CEgKmWUZAZ04cWJ1dTXLMilQa5lXrse8fOeTTeDg1jn3/wyN2EZSCXDgHGitpqenl5eXPaVSBi0YZH76UOWHpc/8BO5FR4CKkb8crSzBTNrrxQAQRcH9A3Ry4g8IFMYajqkkgKmpqY2NDZZ+xhpAcT8V7ZGE8gA7veO7xoJUotvtSinyxMZHOTicRXCBH1hrpRTshM+MdQAWwBhnjBvFdR76+g9LBybosPfV/SVwPqThEvFZPDs722w2Y2sA4AkYNU+MnlkGHjXVRunILx1L2WECMBd8BlY+6X4q1p0auLF2fHy80WgcSO97fPixkwMQArLMRlFEI0CrHO7UB+e03MjkrywsLFy/fl0rmRkrEJQSSrFXB8m5JyPBch0iTwa+z3QOVJ48y0oIRAh8r1yrLi0tMcTBBViepA3/CdEzy8D36TNwgMGM0M7OzsT0lBuWgcIhynr/u3C5lgFqJYTv+5ZcnCaDAj3DVPgnEDYsEHZ2dqamphAHJaC4ovLDCkn+fGYy9r2Oj4/v7e3RMPIsy2wcp/wx67LRGkNPgO6eDHyYeECDxzcSq+EAFhYWtre3HQABOTcoCfZJjvdJ0DPLwPeXwLlliCQ4J7DZbIxVq/CQtdkHV7OWgBQiESnlpanJ7CAHlbjAANBdwgYeXzgEO362NjfrtapAMMaQc/2kZx++vl9ehUMKaS1pracnJnd399qN/dXbq+++++4HH3zQaDTuBRB+AiQAhBsUo7ecVziiGR1sYDz6c56vMIAQx6o1a22702Z/mHWfSiH+u3LcoxcMfHaMgZxomPqfp/jnb4wxRKS1StNUa42IaZoopbY31yfrNTCZkNJlDrQEgQBoTKY8CSNyLN+1A3XOkQPrKe04tNbR1Mz87dWNF8+f6cdZGGjnXBLHQeABkLNWCAGIxrg8UuVeU7jXX9mVkmaDjqpZllmbAkCSpdVSsdtqNnZ3Ws3m5ub60tLSd77znRMnThARlyvhRUiSxPf9I5elYQCcEKLb7RNRo7ff6XTiOO52Wlc+3HPGdrvd/f397e3tbqf10ksvlctla0kIGnXwDPOEHprulK5uqKI7RE5ZyLIk8LVzfFbm3HeIh4fl7xAQQIph5jMogDgzJ4+dWFteK58544QAAWKoGtGwsU5u23+yxxMnVtDB0hERItHDh5Y/gwzMpJQa7XCVh2dmWba7u6uU2tnZCcOQl8/XXq1abjX3i8Wi53mcjcwF9PLG4ndeP99zecmu0NOVcu3K5atnnz+NQly9cj0z/c3V9fGJ+vz8fKlU4u/6QQQAaWa1viNYEwA+zmbm0rmdTscYkyTJ/v5+r9fjpsrdbhcRS6VSu92uVCofffSRtfbs2bM5fuuc833/Phv00qVL29vbXJ2bpbGUslKpyFDW63Wt9cTERJIkV65cOXPmTLVa5eNACNHtdguFgtb6rgdETnfeemCuS3mkt7sQIrNGSJVa6w3XWQihBfPYHYbJHXGjxP8xDwN6UtXr9Vu3bhljtFZCgANwxgk8aEAN9/Y+PgYaxJCLu8nb0eD8h6BnloF5BydJopTK+01euXKl3+9LKUulkjGm1Wo1m812u80imt+89NJLnEHm+/4A9hw+zgPFGxFHQsQY1AEA56BUCAqFwvr6+vLyUuT51VrxhRdeaDT3Ll26ND09PTMzE4Yh18rV+lEWnwP6lFKXLl1aWlqKokhrzU2Sp6amFhcXS6USIu7u7pZKJd/3+/3+6urq5ORknl5zL+7lX166dClN07Nnz/q+XyqVKpWK7/tBEBhjrLW7u7txHNfr9fHx8du3bxeLRd/3WRnJmfYRdj/HP+bgMMel8kwJQLP2ZB0rUNaStVZ78iA5aqSa350zGn4GnHOFyC8Uwu3t7fn5GQBIMxNolX/oKcxV+Fh6ZhmYa8212+39/f39/f1Wq8W9wo8fPx5F0e7ubrFYDMNwenq6VCpJKXmDEtHGxsbq6urs7CwLhNFNMHo281+dc8aYvMM4gbCW+v3++++/f/7c2fmZ6Y2Njd39PXLGWruysmKMWVhY8IMIB9U26c482/tTzoHNZlNK+eKLL9brdSEEaxMswZrNJrNcsVicn59fWVnRWler1dGaYUdR9OGP9XqdLwvDDd1sNnd3d9M07Xa7/X7f8zyl1Ozs7NLS0uXLl59//nmtdc5+eePIe9G9zo48Xy/LMs4J5xgYa5xQwhGwjhNFERFxp5uj12SnEUePHmZFGskPO3bs2G9+8+H09LRWqLXCoZ8pTVO+7wM9hqeGnlkGjuNYa72+vt7v97nt6JkzZ3zfHxsbE0L0er0sy6ampjzPk1ImSUJEQRAkSTI2NmaM2dnZKZfLnNbD2330eM5NvrxUCG/cOMlW1zaazaanxO7ubr1S7na7Sdw3xpw9e3Z1dfX27dtJkpw+cw4R4yQLw+Bh54WIWZZ5nheGoe/79XodANrt9vb2dq/X01rHcUxEPB5jTLVadc6trKxEUcQ1PdnuutdODcNwd3f3xo0bSZLwieB5XrlcrlQqhUJhb2+PdV1r7bFjx27evLm8vHzq1ClEZM35AQ3gXJGBkbAQHjNr+DxH4ywLYY6d6ne6YJ0Fy77uO5YGAAAPNWrmXAQutSMRUQBMjtWvempnZ2tqasoYk6RZliUA0Ov1arUaP/HPED2zDMwbvVKp8IOZmppCRBYU/X4/iqKVlRXO1eajXUoZx3EQBMwDe3t7m5ub58+f544tnMg2GuPOX2QkjGV7mqZXrny002gdPzb/9i9+Pj01aa0tFArNxt4rr7wiBXDXiN3d3ffee292dnZyaoaji+EOuXR/ECUH0rvd7tLSkhDC87x6vV6pVCqVSpqm1lruNcMaRKVSEUJcvXr1zJkzbKkeSbIZvVe73S4UCqzqs/kw6CdG1O/3gyBot9tRFDFjP/fcc0tLS8vLy5OTk7kR+yDOniPT4RBrTv9ut9vlcpnFuERBAJ1ur1CIcJjf5/ua+z/deR3n3BEtmOu8A0KaGqUU60ezs7MrKytKeYJc4Gtez263y0rZg4/8aaBnnIGttfPz81pr9nxEUbS8vGyM6Xa73W43F1PMtPweAIIgmJycRMQPP/zw7NmzjPfmhi4T8zNj3QCQJMnS0lK73T575vTOzs7JkydrtdrOzg4Rvfzyy2w7l8vlMCoqvXTz5k1jjB9EABVPS77OgytvRJQkibW2Xq8fO3YsCIIgCKy1SZKkaYrDRGWGtbIsY9k7MTFx8+bN559/vtlsViqVe/mHhRDc48Jay9hYo9Ho9/tZlrVaLSlloVDg+jIsFQchip43PT3NNsjDci9XnOLUgps3b964cePChQuTk5PW2na7vba+iYiLi4tSyrGxsaF/3h2pDuucc8445ziGfZh8j25IUknrsm6n3+12C0GohfSVVMr3vQHGsbe39+DDfnroWWbgOI7DMCwWi4VCAQA4rJexlqtXr1ar1TiOC4VCmqb9fr/X67GGubW1devWrV6vBwCVSuXDDz8sl8tjY2PFYlEpleNYLGoAgMXv0tLStWvX3njjjTjJoih66aWXOp0WoFer1VhCSolpkgDAwsJCGIY3b968cuXKyZMnpybH70yuuD8ze57HGgFzcrPZZAs/j1JitDZN02KxqLXudrtRFCFiv9+/cePG2bNn2Yt210VLkoTPnSAIENH3/TAMZ2Zm2BH185//PMuyer3Oi8Aa7+TkZLvdLpVKxWKRlfMH0SBy+c8KOY+W2+W0Wq0oijiwvNVqTU9Pl8vFOE7Pnj2DCGmWetrLL8W3yw8CXofcfRjHcb/fT9NUeRoRraHZ2RlrXbFYjKIgy6zAg3Sx0Qifz4ox/MwyMAB0u92VlZW5uTnf91dWViYnJ6enp2/durW3txdFEQBcvXr18uXL29vbQRCw6cg79cSJE+VymRU5Y8y1a9d+/etfVyqVUqk0OzvLbMxnPG+Xbrd769athYUFPjWUEBurK2NjY16otRJDRdQp7Tdbu+yMUUptb2//8u2fs7RhkciG3+gURndSvr24RIW1ttls7u3tFQqF5557Lm8lxZZ8s9nMXbIM/IRhePz48ffeey/HmfjiuStSCMGAHHdOZ36AEehuf3+f0aY0Tdl84PFMTk6ur68zdDxqYtyLcuCA79vtdt99990wDE+cOFEoFL7whS9cv379+vXr/X5/YmLi1VdftdZmJvMDDUDWOaRB3Zw4jrWWUkrnbH7l/FxwzjkHWutms8nIHCKiFAiglBDSs84KCUoorriW9wy418o/GkuTcyDAkctxB2IPWJ4QMiQBAu4RHHafA+WZZWDmkMuXLwNArVa7evXq/v7+5OTk888/DwCc6cpqYRiGQRCwG4YNP4aX2RUMABcuXGA021p7+/btfr9/8uRJImJJyE+i1+stLS2xKZim6fT0dLlc3t3ZKhYjvhrb3iw2fd+fmJhgOPfq1av9fv/UqVPW2twA41sPkWp7RMfmGp1BEFQqlePHjwOAEIK16E6nw9KYN2Ucx6NrgiN0rzA15oG9vb1Op8M6OQAw4g0AvV6PYTDG8Aa9Ah/ScZrr2HxeVCqVCxcuLC0tbW1tzc3NKaW4ac65c+f+5V/+pVobGxsbkwo7nQ4fH87YLMucc6VSKU1j3twc02KMSdM0iiLGk1kNybKs0+kw2kfDqh0AkPd/Ym8/H0l86jF691jcwigEoBNcdSEvMjMSsHEg9geFru9WwffeA3hmGThJko2NjX/913/d3Nw8c+bMV77yFYY32dbNgWg2Dmu1GiLu7OzEcTw2NhZFUaFQCIKAVc1+vx+Goed5jUYjy7KNjY2///u///a3v827jUMp2KL+zW9+E0XR4uJitVKy1hSLxVxKM5PzvkfEMAzZnZskye7u7t7e3uuvvw4APDytdf4gj/TXZqnFFv7GxkYYhjyLLMv49GEHEpcWyb+Fh+lei8abuNVqrays1Go1BuQZA7PW8jSZT2CkKlgude91LhyhfAAs86WUk5OTExMTjKInSXL+/Pl///d/n5ubm5qaWl9babcaPOtKpdJ1rt/va62tzazN2OnN+jOPp1QqMbIIw7MvCAJ+9ILjMAEGOZbo8pMoyzK+7GjfD3gsujQRIBzJSr3zmszVAPSwYfPPLAMjYq1WYzW40WiUSiU2q3J1MX9aDHLU6/VCoSCEuH379m9+85tarXbq1Kkoiph72dxlG69cLrdarbfffpudq0EQCCFKpZLneVtbW4zxCiE4YomRZ+6omCQJszSD4czGFy5cWF5eXlpa+vGPf/zmm2+WSiW2YEfV0YOTm4h5iVmIrVO+FH/RWut5XqfTuXz5cu6SGV2T+3Mvc8LMzMypU6dYJebTh+38vMAgc8ioL+qhELj8PV85L/6IiHx4vfPOO5VKJQzD06dPb29vs5ceEdnaV0rVarVqtbq5uVkoFPhwZB2HMbz8+rkUZcWBRuIlcRB444QQSZIwTsE8zMFqR4Z9r4C8B5ovHVQP47Wiw0FBAISIgypOD3mLZ5aBnXPFYpGlEz8SxnJy3Jg9wHEcJ0nCUAciMr/t7+/fvHnz+vXrX//613PkM49AfP7551dXV9nVVC6X+VK8v0ulEpC9/OFvgiDo9/vVapUxJM/zWBf1PO+IGJRSnjp1Kk3TZrP5i1/84uTJk1NTU/wnYwx3BoAR7oVhcRkeLUtdz/O63S7r5/1+P47jXq+nlCoWi/dan1FpOWrsKaX29vaWl5d5Wfhf7iNXLpdzRxEz9pHcw5zuI7hyzsnRptw64OPy2rVrxpjz58+vr6+zb4nDVAqFApvu+/v7V69effPNr3a73d3d3Xq9nqvxWZaFYQERuWkbouOCod1uV0oEILaW82FLiUmS7u3t9ft9Imq321JKNkZ4hfNmtw+z9e6yIPenfD88QtbaM8vARDQ9Pe37fqvVyrIsCAJ2HRUKBfavWGv7/f7e3l6r1SqXy0KIyclJhmeIaH5+fmtr60c/+tFXvvKV6elp3u5xHO/u7k5OTr788ss/+tGPWq0WH95hGE5NTa2trdVqtd2drZMnT/7qV786duzY3NwcAFSrVbZ72TE7mmIRx3GpVEqS5Ny5c2tra5ubm9euXUuSZGZmhtWBUTbI5TBHdBNRq9W6cuXKxsZGrkPyVmYv7n0Cku+1Ygxrj4+PSylXV1eTJJmenu73+1tbW7mjiG1jOKxnwohIuT/lGtCR42NYsE4uLi7+9Kc/5UOKIyv41INhpEelUomi6OrVq4uLi0tLS5VKhZUFFsI5pGytVUogou/7cRzz0SOG9YCcc4yDrK0NVm92dpafL5/1s7OzOUqXc/IDFuU6pPUctoERkcgR0JHzLVeh4SHjAp5ZBmYL9vjx4+xG4pZLeVdhlrRbW1uXL1/+4IMPAOB73/seQ7XsNErTdGpqKsuy7e3tRqMxOTlZLpfjOI7j+Pbt27xvuNJSpVJRSs3Nze3s7GxsbJRLhY2NjVMnn7t06RJvEWaqnIFzIcapCMyoSZJ0Op25ubkgCH76058CwOzsLNvJMPL8ciWQ4/5qtVq9Xl9dXe31eseOHWs2myx+jx8/Xq/XkyTpdru8GqMydmAA3gPt5EHy9bvd7tbWllJqenqazcter4eIjKLTMID5AVk3vwW/yfk/Z11rbRAERDQ5OfnRRx+dO3duf3+/Wq4IFDweVoM9T42P19fX1zkqlm0cRGRIjFUPfsS5tRvHMSPMDDVnaZokCa9Vnqe1urrK1WcnJydnZmbytAqeI7stHoGctSjBgTu08nfwJxe9BWcHnbgOq2n30aufWQaO47harX73u9/NsixJklu3brHtqpRi+4q1zUajcePGjXa7ffHiRQ4eLBaL5XL5gw8+YIAkiqJKpbK1tfXjH/+YIRYAmJqaOnHiBKe21Gq1JEmq1epLL730/vvvf/D+excvXrx9+/bp06dvLd0GgHa7ff78eUY7c8nDGpoQotVqeZ63tLRUq9W01mtray+88MLt27c9zxsfHy+VSvn+hhFFmvcro8RKqYmJiUajobVeXFxMkqRWq62trfV6vVwbZ8pF5X3WjXUTIcTFixcLhQIiJkmCiBsbGwDQ7/d55LlF4O5owwn33XCsg8CIHpsD2mzRhGF49uzZH/7wh+fOnWMXXaVScc7t7e01Ws1ut1ssRs655557Lk3TWm1sfX21XK5KiezzCgI/j3JVShHZMCxwHhhDfd1uNx66A4QQ1rm1tbVSqbS6ulooFGZnZ7MsYwiTs7t6vR4fvqw4POQ2dJnrAlqLaUY9hxlASGhJWiQHAE5YB4bAEMYGWwIckI+ghsniOOjgc296ZhmY3TDPP/98t9tlDdYY0263GTTu9/scj8WxHPV6/fr16xcvXnTOzc/PCyGOHz++ubm5vb09OTlZrVar1SrLH611rVaLosgYs7m5GccxCwcOeHr11VfDMHz77bdPnz7d7fWnp6e73e76+noURVNTUxyViYO8/4FlmyTJ2toa75uVlRUefK1W+/Wvf/2d73yHd+Eok+Qhn/1+v1arWWtfe+21Wq3GHhR2DrO6wTc6lGsxQnyjXBQzWzJkwMkJm5ubjF0xArS/vz8/P89imYVDjrTx3HkdHuTR3JUNmLEZLCiVSi+++OLVq1fPnTsnhGi0mogotYqiaHx8nPXkNE339hrWUqFQarU6tVqFW4sSobXE1gQfFNaSc7C6us7YGy9+4AWdTmd/v8nYR6vVmpmZqdfr1WqVYYWtrS12xbH/Hz4OxxrNLeXF6XQ661tL//Yf/83pWHhZu7+pfGfQonKZicNAJ71YgNChl1HzH/7r/xzImi/GWzvw7bf+8LnnTjoiJfTHJhg+swzMHCKl5CAkXlxWsdrttjEmjmOWJIVCYXx8/MqVK+xSCoKg2Wzu7+9LKc+ePdtutxuNRhRFfORzFOHKyoq19ubNm8wqDJLxVn7hhRcmJyf/+Z//+eTJk2zlHj9+/Gc/+9n58+f7/f78/DxfoVgscuhIo9GYnp5eXV211o6Pj1+9epXVPI4Jm5mZgeGOz2fBPk/e60mSXL9+naU0G4G7u7sXLlzgHcybiUYA7SNOqZxwGM/knIuiiM3+QqHA6UdLS0s3b97kEGh2rubeLBiCvfmR9AhQLROfF5xbMj4+fuPGjb29PcbS19bWPM/b29vjoNfTp09fv36zVCotLS1duHBhf3+f1aXcgcQkBLDuEoaFXq/HBgXbVmxu8FzK5XK1Wo2iiIEDVrk5toy9DPAAKHSOeI9iclNTU1NzU5evvV2bCIsVr+RNZ65jbSal6PZ7URiSoSTtzS/MxF0XeGL99tJrL/7e+ERdCjSW3cWfVwaGIYqbn4sAoJTipHMGgfI4Ozb8WNYxmMx+XTb2+v3+9evXfd9nhLZQKERRVCwWf/GLXzDuxSc07z+t9djY2Ouvv763t8flATY2NhYXF2/dusW3O3nyJIcfbW1txXFcq9W2trYmJia4tdKFCxc4aWlycnJvb29+fv5OAxgR2dKLoujcuXM8WcarV1dX3333XY7rYuGJI4FWI8Uf7rIX8yPAOVcoFFicXr9+/cSJE7VabWFhgfmZPzxa2JE9W2KYBoQfF0p5V2KFlv8VQrAhw24559z6+jo7pRcWFrIsa7fbMzNTe3t7MzNTa2srZ86cieOe53n9/iBolMcFw6JmYeiPj4+z0cuDZ9iSdRk+7CqVCvvAO52OtZZ1rpx7R2d3/4nw9BmqLBUq33jz9zPbu7X2Qakmeu1OoebFxnhKkSUitNYACAQRBsXN1db5s6++cOGlarkOIEYbn34eQSxe8SNIKfMz+yQ4RWF8fPz69etcg4KfJR/kLG2WlpbW19c3NjaiKJqbm4vjeHNzc3d3FwD29/eDIFhbW8tLc+SSLYqi06dPNxqNa9euAcDZs2c5m//27dv7+/tpmi4uLt6+fbtWqx0/fvxXv/pVs9nkKIUPP/yQIyUuXLiwvr7Oh86Rh5dDzQxl8S/5iOHc5ldeeWV6enpjY6Pf77Ncyr/I7HGvXEIGY9njyjFh77///i9/+ctut/vCCy+88cYbaZouLy+PoKk0mlZJv0W8IQwLFfBKskt2bm7u8uXLtVrtypUrURQlSbKwsLCxsREEwa1bt1588UXGLNggCkPfWislZlmmtWTuzRmJczAYy8yj3EulEicwzM/Pl8tlvpS1ln31g8gtIhYDPMg84PxeU8g/OcQ4MBClP/zWf/9//ah/c+X9sZlqt7NXrEStbiPQHmUkSJNV1qqd9caFM298+bVvlIIxAOEoPzKQU9Y+dwycB9zC0PXC73HYLTIMw4mJiZdeeuny5cu9Xm9iYoJz4tnsHBsbs9a+++67f/EXf5Fl2alTpxYXF1dXV3/wgx90u92vfe1r3//+969fv14qlcIwZGWYZRfL8zRNOfF4ZWVldXX1C1/4wuXLl48dO7a/v7+0tJRl2cmTJwHgypUrUsqJiYnr169LKTc2NiYmJhYWFjY3N9vt9uLiohupe8qPMA/3BQAOKXPONZvNt99+e3Fx8Rvf+MbFixcZDGPtl7XrUUN39Dej2jUMAWE+ktjNdvz48SOZ7oxy8fLyJnODPm/3LAP6IJQPII/uyrGJEydOtNvtZrOZJEkURdvb28ViMUn609OTN27cmJ2dbrebSlcQjFQYx3GxFCnpATg+lLhUFrM0x8xmWcaPaW5ujnHsZrOZFxVhNS0v5zCa9XF/7h2dS/6t9P9r71qf5Kiu++/c293TM7M7I2kfWq2QtBZaryRLFgKbYAQGgYkdxdgpGz8q+cgHV7nKZf8H8ieXXf4Sf3GVbULhqlS5bJOAE0dFwAaHpAjBMUgRKwRIBu2K1b7Y50zPTHffe/LhTPf2vsSuDEY77Kmt1m6rp6fv7Xvuef9OFJPrnLj3S088S38cfqnc0d4IgpyTJ+Y4MsVcezUwQcP233jTHbffV9TbAMfYZmseQBm7FFUiQxYtzMA2QaVMAauQyUOU/V7qFg4fPhwEwcGDB6214+PjAAqFgsQetdZiD7e1tU1MTJRKpQcffHDv3r1SftTf33/XXXcFQSAxhkaj4XmeKF2iVba3t994442+71++fHnPnj2Tk5MdHR3T09NDQ0Pbtm2TJPunnnrq8OHDBw8efPHFF7u7u/ft2zc8PCyme0dHBy9rrSrcFUWR7BqSpjI7O3v06FHxV83NzckYl5RPyfDTvKXlRX/peXENiBwuFouCNCQbhxzF1EyVDmYWRTe91TXIYcog4zUaDUkI7e7uHhsbi+P4ypUrogMXi8Xp6emPfvSQ5KV2dXV1dG6V7Xh6eloi/PlCznUZsEQqdR8ATcVeBLKsAbEyPM+T3UomTdhYJjYre9VVSz6zVklG5SbP9RtRrN38pz/5+cf+fW5k/NWtXb7iiGFdygVziAJ/V8/Afcc/56NsWGlSShFYZSFLr2Z7f/vb317vXG8IElUTSe1O9ry8CREast12dXVJQU+9XpfQhfhyPc/r7u7O5XK5XIlVnhIAABSJSURBVG5yctL3/R07dkhy36FDh44ePSpgVKlVKV8kTq/UFOzq6iKimZkZQRcQgXb69OmBgYE4jj/0oQ8x86uvvtrX17dly5ZLly719vY2Go3du3dLaUS6MlJEAWYOgmBycrJcLvf09FhrR0dHBchOnHCe5w0PD8/OzhYKhd7eXp3p1SqpI9kkqvT+zDw/Pz82NtbR0SHl0OPj42fPnm1vb9++fbsI4bGxMd/3t2/fLmWSWbU5l8uJu/vatOg0PIZEnRYhGcd2fHxi27atSqmenp5yudzb2+P7fmdnZ6Hob922pb2tPYoj388NDw8HQcBsS6X2tN5Q7idwO0SwltMkNqWcKIpzOU9iRQI8Ilno6UZm0i5Ka6iySgeSbrvWWqXAIK08gvvhfQMz8zNjo28pRYocT7XVK/6u7QfuPna/75QUXGOMIleRQ5Q02Ftan2ST4SSw9dcw1xuC0kBfduNMf0ltLa31wMDA22+/LeJFEicloyCKou7u7q9+9aulUukHP/hBrVa77bbburq6Dh06JInT4gyT4EdWrEkVQVrDUK/Xe3t7C4XCxYsXRW7EcXzgwIE//OEPEljO5/OHDh0aGxsLw/DgwYNDQ0O9vb29vb1LvEHCaTZTcCs2cC6XK5fLzz77bHd3NwAJYEjsdPfu3dlcRckTTKMpy71NEm8T8e66rlih/f39Yh3IpO3YsaOzs5MSZB+RNqVSiTP5WOuPlyLlGVHOJenKGNPR0TE8PLxr187UKBUdpxHWxF0/NTUVho3h4SHJ9xT3JJrFPYlI5CaMe+qLstYCKpfzrDXlchmZHSRV5lMP6NqHIHdIP6W1juNQkSboOIy1V773js//5j/Ci0Mvd23bOjVe7dt18713fa6gtwJOI6z4XiGOoRyVTmM2SL7il66pfGQjkiA/rm4/NCkdvvgqhEMajUYQBCItATQajTiORaeVvKvU6ZrSat+Sxl0FQ+PUqVNhGO7du1fMXdFRBeqxXC4LYE1bW9vAwIA8zJINSPjNGDM2NnbmzJldu3YJi2qtT506deTIESmQYuaXX35ZMooEwyCNJ4sWmrJHOmTJBhsZGTl79qzkeAGQiKggXeXz+UqlMjs7S0Q7d+4slUqSoApAyiRTCXyV2VgjZddkFJnJycnZ2Wl5BiKKooa1NqhVZOCe55VK7b7vB0EwNTUl/qdyuUxEzM1OK5atIpFVS3eW1db/nziEDFlAxRG0FM6gUrNX/vO/n/rfF35/68fvvPWWT23J72B4sKyUVCO7q4O/r/ScrcrA4qlasvqxSiXX8vPixbEJinI2xCd/Zm2eq1D2s3Nzc8x86tSpYrF42223DQ4Oinm5ZcuWnp4eCU7u3r17z549EiumVaB2arXaxMTE4OBgEAQ33HCDmOtS2JBGYsbHxwUrQ4AywjCUTIlKpSLlssjs6ykDj46Ovvnmm+fOnSsUClLY1Gg0BLZWkkbL5fL8/LzneceOHZMSjtnZ2cnJSdFWUiH5LjIwEUWRCcN6mskMWPFXO+7C7qOIGM1dMp/PN9tiJOUBDCaIBPvzA8daQJkYWiOMjeHA98xUMDE5Nr69u7dc7AVcK7aRiphZ0WqYnpsMvHqcI9Unl/9XWvq3xu9dvnzT702Dz+fOnRP9Vpwuvu+PjY0FQbB3796Ojg45cxU2kHKC8+fPDw4OStJvEAQSCJFVLlHNarVaKpVuv/329IMjIyOScS1R0NTXkjKwgE4+99xzEvUVqSuAu1J73NfX9/rrrzcajbvvvluuj+O4WCxu3bpVInNL0iSvjZYwcNZnnujn2dVs047qy+6T3WTfP8xnRhzDcSXxrub7HoPjOHIcF0AcWUBdFc0zO9ilo2hZG3g5Xd2FuOSCxEyCqKyU1Nav5W7LI7fMLMn0cRyXSqVbb7310qVL586dy+fzhUJhcnKys7PzyJEjEn29enMDUQrm5+cPHDgwNjZ24MABUe8lL1pSKavV6s6dO4eHh6MompmZaWtrGx0dFShsiRVnfLMLY5EIeRiGoj9XKpVqtSoRVGbu6OjwfX/Pnj1XrlwJgkCgwrTWPT09UugnsaWrBFrWTtnNNLX5UxuVma1lOYema0CDbGrUM4z0Lnn31OBrJ2ugFJSWtQTfL7BFrRblcgU2IMXaicDEbMCaCOstKGxlCZzakGu5fl2K35JJW8L5WIm9RcoJYosULQpuhkSMpWwgvezqzxyGYRAElUrl8uXLqdkpLCQcWKlU+vv7z5w5Mzc3d+DAAQFq7evrk68WGz4LvpUqIKK2pFMhVXhEVK/XJXWMiCQGtnv3bmFdcTtloeTeI8q+oDSlDAtTbZOWJYrZpDUA67J33guKIzgOQNbYiIgUObCKLZrqgrKAscxsSZHbfMZlayfz+9KF0bIMLBHL5TbkaoyaOlHlz/SVi3NrOUctn7e1LBFB0kibGAgbC2hmNuoofJ59mPQOclk2KVKqhcS+tdbOz8+Pjo6WSqU333xTKbV///7e3l5xqqc+2KsobOkWkyJyUJLClfq9lVJyjUpw8LIZSLwMB/9PoawDNpuQnH2PV+HSd8xhfs+p2SHRMmICMSuCblb+AhaR0kxQgEo6J62ZgaUXVKsycBo1Xc0ivbbbXhvfArCZJiwiBlOBXKvVpD4ZaysZtwmYY+pPxuKtQTJ+Bb9e7rbk5lldlxdnZclxuTKcPZPuL9mxS+7HtbUmXDulX2ot0nnilfqEJn2e39PHWRNZCyBWiVuNjQKaHSRA0sxVGBjGQC+a9eWOq3TMzX9bgIFlkGrJmYRRs5Aoi6aDiLKfynhfVqj/eLd8M+LKEuClNC9qCf+sJY6axpPEb5SVS9k6ofRiJKKMVo8Ap3HsNLU7fR4Zu1jm6ZUibFN1IDUB/vSJWj7YbApdVtgu4dI4to6TovwsnM9y+/KbJ7/K1c2lkkzOko+tu3ugKHZEDFhrjNJJnnbSPJWttUyrWB+rMzASJ/sGZ2ALSLKrs9DNGTFgCGAQW6VUc0+Lo5rjOoBNggoURcZ1c8aw1qkdhcUvaeO1q9ukdZJs96qpkSowLMOwNUppggZUlgmb0nIVytZvSzqt1H66rgvYBGNgwWOazdC8SmvlqzhoWnKBMmAbUY1AoplEIQNQjqo15i1iZjbWyJtgBpgSYHC1+PW05ORs0hJStVqNCJYZxICt1auUoEaLzU9JC6s4MqutiqwJI06Ter0uCaECkBCGseN4UWTEYZH2Xk7vIHrZegXqRpfAWAS4SwAsEAGWoWJDjvYAWAvLsaMNwQa1eiHfHoax5/m8Uo+sTfpgEQMEYyNSMRFXqpX2YjmMDFudy+XAiMLQFY99zHDeGTcy9WhIsbFNSBwEzCxFL8ws0AVpUk1qF6zLQdNCQiYJMQAAFKAd7YExPxcqBaV5rjox+vbQv556fPLtMZAFLAONhllkBGV/NumDQVEYa20VRa9dPP3kU7+amZsgsq6rJXTvOI41BoC1BCwWGItJmtTm83mRpdVqdWRkRLorp46PoaGh733ve2+88Yao1tmUuGwtx9ofvoUSORiglHuJ2bEWUPCLnkVcjd5u8NSjv/p5WCVyIu3GMSpa5VyfElcIUgdGc1+T7bCV9rhNWkoqjuF4ZLhea0yxql289HLtifkvfuFvHaXrtYZSRa1dG1m2UC5sDLUKx9gEDl4SWpRSL7300je+8Y0f/vCHFy5cGBkZOXnyJDPv2bOnvb39lVde6evrE2+/6NtriT6sSBufgcmu5HZS0kLWMJS2Qfz2XOWtRx97uFgsMjtwG4wgtJxXYCgCxTCJ91m4XllAYUU/5Ca1FGnXAWKFuOi7FlXthzPVkV+d+tldd36mZ+uu2IZkSbtO1Ihd12GsqpoJE6awXgAOHjzY09Nz7Nixrq6uzs5OAK+99torr7zy/PPP33zzzZOTky+88MLExERvb++dd97529/+tlar3X///ettULzRGbhp8QIOmLLD0RpRZFhHQTRTqY38/LGflLdp3zfG8Mz8eKHgeipf45kotq72SCkCGJaaNpFK8+DFJfZ+DW+T3muar8xua9vSQCXmoBrOtW9xC0U1MfXaf/1P9Knjn23L9dQbUd5rdzwNLAnSLiKbtEeTP5lZAAYfffTRp59++sc//vHQ0NB3vvOdhx566MyZM/Pz89/97ncffPDBZ555Zv/+/T/96U9vuumm6enpH/3oR9/61rewnrzAFmBguyKDMThCJadMpTbyi8ceat8GLx/Xw6n5evTE078wMfL5YhSy1q7WWhwPi0LKafs4Mtk0vc1jix21Q2EYak2OS0FjjlSkbKPc7V0eP/fEM7VP3/Olcq6nHs7lvXZrBWlg5UhPmnwmKYBSJV4qlb785S8LyP7g4OD27dsl4bxQKNxxxx2//vWvXdcdGBj4/ve/L0XdH/vYx9bLABudgdOoDwGClQ0oEFkgIGd+onL5Z7/8yZbOnOtzGFUcX3fd4ILrzIptpAsKUETU3ibeeLtwW6bma95k4FY+QpPjgxgEGN2WAzRTPUKj+4bC8OXBf/uNue/433Tm+xpc1ZRzlLdaJp+ckcQ4OQo+IREdPXr05MmTX/va1x5++GHpiTk7OwvAWnvixAlmPnLkSFdX1/Hjxy9cuLAkJ/QdRfFGDyPZhOuSVFIGk7FUt5gbmXr90cf+oXNnsVqfLpb8OI7DqO77DjMDDtumr3nxBCXpIOkNyTJZEnyHzWNrHQEYw2AFKCYQMWujYADLTDZ0opqzrX3PXbd/rru9D9YnOJqcZWsGyKSaSpQoDMPTp0//7ne/27FjR6lUmpqa+spXvvLiiy8ODg6WSiXf90dHR6Vx3JkzZx566KFHHnmkXC6fOHFCOt2uPZK00Rl4mVuf2aAWY25y+o+//JeHy9vIONVie362Mp/zClCITQCy8s6aXNr0XdvMXVQWVcwSFG8eW/AIQCkCWWa2kJIDwaFiQDnwEDrVOdue67v/03+31d/daCDnrczASKJBaXZ6EATSkCXtiaESqONqtfrNb37z61//utb68ccfP3nyJBLpvd7qi5Zg4FRqAsxxZKsxzz/yj3/vl+NcoRarquE4NHHObw/D0HENyACU8LCE3SwRJZyMBfZezNab1HqkANHjLFkANgGgJVZkKUf5esWJ5vNdWz78hb9+UKscmmmVqzIwkvwtz/NELAuqibiprbUS/q3X64Il3N/fLx4vlXROXBcDb3QbGMCivAsiKHYUubt27rv41untRT8Kq07OyXu5eiMkguU4NS+ABMpYWDcDhQ+YpLwr3uTi1iVtYgekSVlWDJL2SgCUp735Sj1XcAv5tpEr1Vs+2meMIQu1SslzCjMmMF1yUkqvBSeQkrZMkizted7hw4ezJq4ACa4IonQV2vASOCN9F9jMoBGZ6Sef/ecLb/x+a0/Oy9t6XGMox1VxXAdArJiU2EJi7ygwq8TWtfK/ismC4mZoafPYekdWin1AkYrlXVuAmRQrtk7OaYuqNPZW5a/ueeBI/7E4zjs6D6zsxGouvATyHkBaPZqSMLn0Q0XGQSUdquUj7wjDuIRag4EtYABI7YhAmIVxNdazzzz7T4MXn9u+sy1CxXDoujmyDqDJkoUiy6xIwVqyim3GOlLNo3yH2Mybx9Y7ovmumQ2TBRlShpQBOzCOMsXxy8E9d95/dP8dbHKubrMGpFa1gVPrN0s26SwpcSbJqYyiSIDW0gSstP3AupBh0BIqdNMRzWC2FlCiUHtOMY7CT95xIuLGpZH/K5ZyhXwujkytArYkLmhuIic1O0EvCiMtHDepxUnar5CjlLbKdR3XdZUbh3pionr3sRNH9t+mUWDlWmOUdlaTd6IVp32nOOnVBixUR0dRlPaXlDNyTbYSe70JlRteAmcoiScBAAxbRkgUVqIrzz3/5OtvnN6yNU+kpifNkUO3KqXz+aKJ2Rj2/bwxhm1m22uaxGrh9+uGaDF6ZjZguOL1q52X7V/sMcdxpCXFu/601z9prZmt0lEjrkxNXzl/4ZXOzk4wjY3O3Hf8s/v33aKRd1Fg6CiMXNddHfb1/aGN/84yRjCANECvSEUxacctudtv//inFKuz51/Y0bOr4NCBfbds3dLhwEMzgu8wWEEnzJ/lWF6HNvPnJRZ/aOJ5U+vUFxhs2SpSsYkd3SyTM2z0O7WEbzGK2CjioDHdlnPPXnzx0h/HTD03PTX/l/d8sW/3gIdSzMSkCcpxXaLrjl+uuwe6FlqJyayF1q6CtYhLfvftnziulDp//ryJ8p5bcJCL2ChyJOlKk5Ph2iwbX3cMvIClKlkoC1WU69OkFBQRCEprx1jrKCe2saMc+8FyuSuXfMNRW25rhArBDxs0Oz37mc+cOLjvJg0fcAmK4ABKZXq1XD+08Rk4XcCLMzGUQhzDEhztG9g2t+sTf3FvWHcuvTHWqFsUHU2OggsoSwCUWng3KSdYrLw5vJ/ESd/nJYtptefkVRmborCR8xwFKAXL1hoF5XzA7H4VRdZ1/UptppAvulSYmqw98MADA/0fYbgGjoKjyWUGM1TG53T9UMvYwHaxC0pFMRyniWzGCGMbuAq1qPby2fP7Bz5SKBQ0uQyihWzq5TcEcP3WA6dwc0LXsLCCIBAc6bSh5vuMwPr+UDPaH3M4X5kaHr50+OBhAxBcgicNWUwErXENqOt/BmoxBuZmGnPCwADC0GqHtWLDoW4iUSqJH4VRrLWrV47NX9cMfM3otillIWYFYlKO15uEea8pjqA1GCAyTIbAACxT6qyS6ITMirEp/uH1Qi37tlyNehDBIucqhzTBIesCDsG1BmxAcHKu7yhNANumOyvzo+Tn/R7HUkqAVBcIIL4WU10BqFQCwUkU/N0PGvcCcDTYQBGIdFCpxxEIroLblAgWxBlI2gT98PqhlnlhMhBKLVjLyOebvSqMgTGstWtiApTWrlR1xnEzg1LJ+t8IukhW0BrDjUZUq9Wr1epq14v9ttIPT05OnTt3rl4PiaAUrEWLaGNrJ0YcQWlYC2vR1lZy3ZwxJApNOtXMMJbZQl9/kbb/B3tGIXmK8LwVAAAAAElFTkSuQmCC"}}},{"cell_type":"markdown","source":"# Conclusion\nOrchestrating timely and appropriate actions for Cyber-physical systems is a computationally intensive process with applications in various domains, including robotics, smart homes, smart cities, autonomous vehicles, and industrial control systems. The goal is to distil a vast amount of information into a single, well-timed, and contextually significant insight.\n\nFog computing emerges as a method to accommodate large models with low-latency needs, capitalizing on the benefits of Multi-Agent Models. With the progression of 5G-V2X standards and the introduction of the new Robot Operating System (ROS) platform, new avenues for applying reinforcement learning in real-world solutions are becoming more accessible. Furthermore, innovative Federated Learning techniques strengthen the case for edge computing.\n\nOne could contend that computational capacity is not keeping pace with the algorithmic possibilities offered by machine learning, particularly for high-volume, low-cost cyber-physical machines. The challenges related to scaling the state space are considerable, but surmounting them paves the way for new innovative solutions and entry into new markets.","metadata":{}},{"cell_type":"markdown","source":"---","metadata":{}},{"cell_type":"markdown","source":"# References\n\n<font color='grey'>\n    All references are used with the notation [#] in the text.\n</font>\n\n<br>\n\n1.\tTimeline for deployment of C-V2X – Update\nhttps://5gaa.org/content/uploads/2019/01/5GAA_White-Paper-CV2X-Roadmap.pdf\n\n2.\tTimeline for deployment of C-V2X – Update\nhttps://5gaa.org/content/uploads/2019/01/5GAA_White-Paper-CV2X-Roadmap.pdf\n\n3.\tMulti-robot path planning based on a deep reinforcement learning DQN algorithm\nhttps://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/trit.2020.0024\n\n4.\tResource Allocation in V2X Communications Based on Multi-Agent Reinforcement Learning with Attention Mechanism \nhttps://www.mdpi.com/2227-7390/10/19/3415\n\n5.\tReinforcement Learning in Robotics: Applications and Real-World Challenges\nhttps://www.mdpi.com/2218-6581/2/3/122\n\n6.\tChatGPT – Query: Define Cyber-physical systems where actions in physical systems are software driven.\n\n7.\tMulti-Agent Fog Computing Model for Healthcare Critical Tasks Management\nhttps://www.mdpi.com/1424-8220/20/7/1853\n\n8.\tFederated Learning in Robotic and Autonomous Systems\nhttps://www.sciencedirect.com/science/article/pii/S187705092101437X\n\n9.\tChatGPT – Query: Explain the advancement from Big Data to Fast Data and Actionable Data\n\n10.\tThe Big Data methodology in computer vision systems\nhttps://ceur-ws.org/Vol-1490/paper51.pdf\n\n11.\tAI and IoT(S): A Haven of Possibilities\nhttps://papers.ssrn.com/sol3/papers.cfm?abstract_id=4158707\n\n12.\tSim2Real Grasp Pose Estimation for Adaptive Robotic Applications\nhttps://arxiv.org/pdf/2211.01048.pdf\n\n13.\tComputation offloading in mobile edge computing networks: A survey\nhttps://www.sciencedirect.com/science/article/abs/pii/S1084804522000327\n\n14.\tResearch on Big Data Mining Application of Internet of Things Based on Artificial Intelligence Technology\nhttps://ieeexplore.ieee.org/document/10086442\n\n15.\tEdge Computing with Artificial Intelligence: A Machine Learning Perspective\nhttps://dl.acm.org/doi/full/10.1145/3555802\n\n16.\tA Hierarchical Federated Learning Incentive Mechanism in UAV-assisted Edge Computing Environment\nhttps://www.sciencedirect.com/science/article/abs/pii/S1570870523001695\n\n17.\tFedDNA: Federated learning using dynamic node alignment\nhttps://journals.plos.org/plosone/article?id=10.1371/journal.pone.0288157\n\n18.\tChatGPT Query “summarise Federated Learning in two or less sentences.”\n\n19.\tAn Edge-AI Based Forecasting Approach for Improving Smart Microgrid Efficiency\nhttps://ieeexplore.ieee.org/document/9744454\n\n20.\tMIT news: In the blink of an eye\nhttps://news.mit.edu/2014/in-the-blink-of-an-eye-0116\n\n21.\tAuditory processing in the human cortex: An intracranial electrophysiology perspective\nhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC5562943/\n\n22.\tWorld Rob\nhttps://ifr.org/img/worldrobotics/Executive_Summary_WR_Industrial_Robots_2022.pdf\n\n23.\tMulti-agent and reinforcement learning based code offloading in mobile fog\nhttps://ieeexplore.ieee.org/abstract/document/7427078\n\n\n","metadata":{}},{"cell_type":"markdown","source":"---\n---","metadata":{}}]}