Signal Processing in the Retina: Interpretable Graph Classifier to Predict Ganglion Cell Responses

• Published in: IEEE open journal of signal processing Vol. 5; pp. 1 - 9
• Main Authors: Parhizkar, Yasaman, Cheung, Gene, Eckford, Andrew W.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.01.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Classifiers; Combinatorial analysis; Computation; Feature extraction; Firing; Ganglia; Graph signal processing; interpretability; Mathematical analysis; Matrices (mathematics); Measurement; metric learning; Neural networks; Retina; retinal ganglion cell encoding; semi-definite programing; Semidefinite programming; Training; Visual observation; Visual stimuli; Visualization
• ISSN: 2644-1322; 2644-1322
• DOI: 10.1109/OJSP.2023.3349111

It is a popular hypothesis in neuroscience that ganglion cells in the retina are activated by selectively detecting visual features in an observed scene. While ganglion cell firings can be predicted via data-trained deep neural nets, the networks remain indecipherable, thus providing little understanding of the cells' underlying operations. To extract knowledge from the cell firings, in this paper we learn an interpretable graph-based classifier from data to predict the firings of ganglion cells in response to visual stimuli. Specifically, we learn a positive semi-definite (PSD) metric matrix <inline-formula><tex-math notation="LaTeX"> M ≥ 0 </tex-math></inline-formula> that defines Mahalanobis distances between graph nodes (visual events) endowed with pre-computed feature vectors; the computed inter-node distances lead to edge weights and a combinatorial graph that is amenable to binary classification. Mathematically, we define the objective of metric matrix <inline-formula><tex-math notation="LaTeX"> \rm{M} </tex-math></inline-formula> optimization using a graph adaptation of large margin nearest neighbor (LMNN), which is rewritten as a semi-definite programming (SDP) problem. We solve it efficiently via a fast approximation called Gershgorin disc perfect alignment (GDPA) linearization. The learned metric matrix <inline-formula><tex-math notation="LaTeX"> \rm{M} </tex-math></inline-formula> provides interpretability: important features are identified along <inline-formula><tex-math notation="LaTeX"> \rm{M} </tex-math></inline-formula>'s diagonal, and their mutual relationships are inferred from off-diagonal terms. Our fast metric learning framework can be applied to other biological systems with pre-chosen features that require interpretation.