The majority of tools for bacteria detection and recognition are based on physiological or genetic properties of microorganisms. However, there is an enormous interest in devising truly label-free and reagentless biosensors which would operate utilizing the biophysical nature of the samples without the need for sensing and reporting biochemistry. The phenotypic biophysical sensors are closest to realizing this goal. Whether they are based on mass spectroscopy or on forward scatter phenotyping, phenotypic biosensors rely on a library of spectral signatures generated for different bacterial classes to subsequently detect and classify future samples of unknown nature. In this competition participants will use unsupervised feature learning and supervised classification techniques to classify bacteria from four different genuses (e.coli, listeria, salmonella, staphylococus). What makes this problem challenging are:
No hand-crafted feature extraction is allowed. All features should be derived in an unsupervised manner starting with the individual pixels.
Training, public, and private data sets were collected during a 3-day time frame.  We speculate that data collected on different days could have been impacted by random effects. Making sure that a classifier trained on day 1 data set generalizes to day 2 and 3 data sets is the main goal of this competition.
Data Set
You will be provided about 1800 images. This image set is split into three as train, public, and private. You will be provided with labels for images in the training set (Train_images.zip) and build a classifier to predict labels for images in the public and private sets. Public and private images are provided in the Test_images.zip file. You won't know which images are public and which are private. When you submit your predictions your public and private scores will be separately computed but you will only get to see your public score.
Acknowledgements
We thank Dr. Bartek Rajwa, Ph.D of Purdue University Bindley Bioscience Center for providing the bacteria dataset.