Context-Learning (cTL) : Deep Learning Framework for Autonomous Context Learning and Switching
This research introduces a new neurobiologically-inspired deep learning framework for autonomous context learning. This framework is compatible with Tensorflow Keras to allow simple integration of context learning/switching mechanisms inspired from working memory research and models into typical neural network architectures. Work reported in: Ludwig, D. W., Remedios, L. W., and Phillips, J. L. (in press). A neurobiologically-inspired deep learning framework for autonomous context learning. In Proceedings of the 33rd IEEE International Conference on Tools with Artificial Intelligence (ICTAI 2021). Bourbakis-Ramamoorthy Best Paper Award. [preprint-PDF].
ElectroStatic Surface Charge (ESSC) Pipeline
The ESSC Pipeline allows for parallel computation and analysis of electrostatic surface charge data across a wide range of protein mutants. The pipeline has been applied to screen for differences between transmitter-founder and chronic-control HIV envelope protein sequences, identify HIV variants transmitted between couples, and explain differences in transmissibility between COVID-19 variants. Work reported in: Howton, J. and Phillips, J. L. (2017). Computational modeling of pH-dependent gp120-CD4 interactions in founder and chronic HIV strains. Proceedings of the 8th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (CSBW), Boston, MA. [PDF] or [PDF]; Morton, S. P., Phillips, J. B., and Phillips, J. L. (2019). The molecular basis of pH-modulated HIV gp120 binding revealed. Evolutionary Bioinformatics. [LINK]; Morton, S. P., Howton, J., and Phillips, J. L. (2018). Sub-class differences of pH-dependent HIV GP120-CD4 interactions. In Proceedings of the 9th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (CSBW), Washington, DC. [PDF] or [preprint-PDF]; Morton, S. P., Phillips, J. B., and Phillips, J. L. (2017). High-throughput structural modeling of the HIV transmission bottleneck. 2017 IEEE International Conference on Bioinformatics and Biomedicine Workshops, Kansas City, MO. [PDF] or [preprint-PDF]; COVID-19 work current under review.
MDSCTK : Molecular Dynamics Spectral Clustering Toolkit
The MDSCTK provides a clean interface for performing spectral clustering on molecular dynamics trajectories. It assumes some basic experience with GROMACS (http://www.gromacs.org) and a little R (http://www.r-project.org/), and uses the ARPACK (http://www.caam.rice.edu/software/ARPACK/) routines for performing sparse eigen decomposition as well as parallel, database-driven (ORACLE Berkeley DB) methods for fast computation of large, sparse RMSD distance matrices. Work reported in: Phillips, J. L., Colvin, M. E., and Newsam, S. (2011). Validating clustering of molecular dynamics simulations using polymer models. BMC Bioinformatics, 12 (1), 445. [PDF]; Syzonenko, I. and Phillips, J. L. (2018). Hybrid spectral/subspace clustering of molecular dynamics simulations. In Proceedings of the 9th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics, Washington, DC. [PDF] or [preprint-PDF]
Cell-Diff : Analytic cell proliferation/differentiation model distribution generator
Another past project focused on the development of a tool to fit parameters for cell proliferation/differentiation models to experimental cell count distribtions. Common methods such as Monte Carlo simulations have trouble with this kind of problem since many cell count observations occur with very low probability. Therefore, I developed a tool to generate C codes for (unsimplified) analytic equations which can be used to numerically calculate the resulting distributions to machine precision and aid model fitting. Work reported in: "Analytic parameter fitting in stochastic stem cell models." J. L. Phillips, J. E. Manilay, and M. E. Colvin Biophysical Journal 98(3), 739a. (2010) doi:10.1016/j.bpj.2009.12.4052
TD-ALCOVE : Temporal Difference Learning of Dimensional Attention in ALCOVE
This project focused on the development of a version of ALCOVE that utilized a more biologically plausible mechanism for learning dimensional attention based on temporal difference learning instead of traditional error-backpropagation learning. Work reported in: Phillips, J. L. and Noelle, D. C. (2004) Reinforcement learning of dimensional attention for categorization. In Proceedings of the 26th Annual Meeting of the Cognitive Science Society. Chicago, IL.
Robot Prefrontal Cortex (PFC) Working Memory Toolkit (WMtk)
One past project focused on the development of biologically inspired computational mechanisms for effective robot learning and control. In particular, David Noelle (Univ. of Calif., Merced), and I developed a software toolkit that allows for the easy integration of a powerful computational neuroscience model of working memory into robotic (or really any artificial) systems. This model of working memory has been used to train robots to perform standard laboratory tests of working memory function, such as the delayed saccade task, as well tasks in robot navigation, motior skill learning, and object manipulation.