Fuzzy AI Lab Projects

[[[[Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an]]]]

Lynn Pickering and Dr. Cohen

Summary: A fuzzy system trained by a genetic algorithm offers explainability and transparency in its decision making. Here, an aggregate fuzzy system works towards explainability while greatly reducing the number of rules needed to describe the system. The genetic algorithm, fuzzy logic and aggregate fuzzy tree are the separate parts that make up this system, and have been summarized. This system is trained on the Breast Cancer Wisconsin Data set. Two variations in the training of the system include the genetic algorithm mutation rate and the structure of the aggregate fuzzy tree. The method with the highest accuracy, where the tree structure is fixed and the mutation is varied, is examined closer to illustrate the level of explainability and transparency of such a system. An accuracy of 94.96% and a sensitivity of 98.08% is achieved on the test data, and while slightly lower than accuracy achieved in previous works on this data set, the model trained here works towards explainability, which is of high importance.

Final Outcome: A book chapter in the 2021 Springer series titled ‘Explainable AI and Other Applications of Fuzzy Technique’s’

Lynn Pickering and Dr. Cohen

Summary: Tetris is a single player game, the objective being to place four-piece blocks and clear as many rows of blocks as possible. The game requires quickness and flexibility in its decision making, which makes it a good candidate for Fuzzy Logic decision making. To train the Fuzzy Logic Tetris Player (FLTP), a genetic algorithm is used. The genetic algorithm trains the input and output membership functions, along with the rules of the fuzzy logic system. Previously the FLTP was created by playing the game repeatedly, creating input functions and rules and iterating upon these parameters by identifying faulty decisions from previous runs. The genetic algorithm is able to search a bigger solution space, exploring this complex problem and the many parameters to tune; therefore, achieving higher game scores than this previous FLTP not trained by a genetic algorithm.

Final Outcome: A book chapter in the 2022 Springer series titled ‘Fuzzy Information Processing 2020’

[[[[Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an]]]]

Summary: The objective was to predict individual Cognitive Processing Therapy and Prolonged Exposure Therapy outcomes for PTSD post-therapy using a genetic fuzzy inference system trained with baseline data only.

Outcome: Developed a GFS that predicts the best course of therapy.

Leveraging Artificial-Intelligence to Profile and Enhance Phenotypic Plasticity for Second Injury Prevention: An Innovative Precision Medicine Platform to Revolutionize Injury Care

Summary: This project is in collaboration with University of North Carolina, Chapel Hills and is funded by the National Institutes of Health (NIH). This involves training fuzzy AI models for predictive analysis for Brain Concussion diagnosis and prognosis. Additionally, fuzzy systems have also been trained to understand the athlete dynamics in soccer. Such a system can then be incorporated in a VR environment to control NPCs to intercept human players. This provides a training environment to limit movements that have higher propensity for injury.

Publications:

• Sathyan, A., Harrison, H.S., Kiefer, A.W., Silva, P.L., MacPherson, R. and Cohen, K., 2019, June. Genetic fuzzy system for anticipating athlete decision making in virtual reality. In International Fuzzy Systems Association World Congress (pp. 578-588). Springer, Cham.
• Kiefer, A.W., Sathyan, A., Reed, C., Walker, G., Elpers, J., Gubanich, P., Cohen, K. and Logan, K., 2020. Predicting Protracted Concussion Recovery To Inform Proactive Care: A Genetic Fuzzy Machine Learning Approach: 2830 Board# 291 May 29 9: 30 AM-11: 00 AM. Medicine & Science in Sports & Exercise, 52(7S), p.785.
• Kiefer, A.W., Armitano-Lago, C., Sathyan, A., Longobardi, L., Loeser, R., Spang, J.T., Cohen, K. and Pietrosimone, B.G., 2021. Predicting Posttraumatic Osteoarthritis Related-symptomology Using Serum Biomarkers: A Novel Explainable Machine Learning Modeling Approach: 364. Medicine & Science in Sports & Exercise, 53(8S), p.114.

Summary: The objective is to predict the time needed by an athlete to return to play following a concussion. This project is in collaboration with Cincinnati Children’s Hospital Medical Center. Data from DTI image scans are used to make the predictions. Principal Component Analysis (PCA) was done for variable reduction. GFS models were trained for regression and classification. For the regression case, the GFS model was trained to predict the return to play time needed whereas for classification, another GFS model was trained to predict if the return to play would be short (≤13 days) or long (>13 days). This pilot study was on a small dataset consisting of 34 datapoints. We are in the process of gathering a larger dataset for a more comprehensive study.

Publications:

• Sathyan, A., Yuan, W., Fleck, D.E., Bonnette, S., Diekfuss, J.A., Martis, M., Gable, A., Myer, G.D., Altaye, M., Dudley, J.A. and Cohen, K., 2021, June. Genetic Fuzzy Methodology to Predict Time to Return to Play from Sports-Related Concussion. In North American Fuzzy Information Processing Society Annual Conference (pp. 380-390). Springer, Cham.

Neurodevelopmental and Clinical Trajectories of Youth at Risk for Bipolar Disorder

Summary: This is a collaborative effort with University of Texas, Austin funded by the NIH. This two-site linked R01 is a large longitudinal study examining the neurodevelopment of youth at familial risk for bipolar disorder, with goals to understand neurobiological risk factors for key outcomes.