An ever-expanding collection of papers and synthesized notes at the intersection of AI, biology, and mathematics. Browse by topic to find domain-specific works, or scroll through the continuous paper feed (at the bottom of this page) for the full collection. This repository is powered by an automated sync system that pulls high-signal resources from my personal Notion database. Please open an issue to report problems or request that a resource be added.

The core "how" of both biological and artificial reasoning.

• information-theory

  The mathematical basis for neural networks and genetic coding.

• stochastic-processes

  Modeling randomness in Markov chains, protein dynamics, and cellular trajectories.

• optimization-theory

  The engine behind AI training and evolutionary biological selection.

• causality

  Causal graphs, AI robustness, and gene regulatory networks.

Turning complex, messy data into computable "maps" of the world.

• embeddings

  Representing words, LLMs, and chemical compounds as vectors in latent space.

• geometric-dl

  Graph neural networks for protein structures and medical knowledge graphs.

• multimodal-integration

  Combining disparate data types like genomic data and clinical EHR notes.

• interpretability

  Mechanistic understanding of deep learning models and cellular pathways.

Modeling change over time, from reasoning paths to life cycles.

• sequential-decision-making

  Reinforcement learning for AI agents and organoid protocol optimization.

• generative-design

  AI-driven "hallucinations" of protein structures or synthetic gene circuits.

• trajectory-inference

  Mapping the "life story" of a cell or the reasoning path of a model.

• perturbation-prediction

  Active learning to predict how systems react to external triggers.

The implementation of theory into clinical and laboratory practice.

• knowledge-synthesis

  Using EHRs and knowledge graphs to bridge bench science and clinical care.

• structural-omics

  Spatial transcriptomics and the physical architecture of protein design.

• automated-discovery

  Using AI agents to design and run experiments (Active Learning).

• robustness-safety

  Predictable cellular reprogramming and counterfactual-proof medical AI.

Topic: foundations-intelligence/information-theory
Citation: McCulloch, Warren S., and Walter Pitts. “A Logical Calculus of the Ideas Immanent in Nervous Activity.” The Bulletin of Mathematical Biophysics 5, no. 4 (1943): 115–33. https://doi.org/10.1007/BF02478259.

Summary: At the time, McCulloch and Pitts inherited many important findings about neurophysiology: neurons were discrete cells, nervous activity had an all-or-none character, synapses could excite or inhibit, and signals unfolded in time. However, there was no formal computational theory explaining how these biological mechanisms could produce thought, memory, or inference. So, the authors developed a logical model of neural activity, treating neurons as binary threshold devices and neural nets as systems of propositions unfolding over time, which solves the problem of connecting neurophysiology to computation. Their model showed how physical nervous structures could, in principle, realize logical operations. They discovered that idealized neural nets, when modeled through Boolean logic, could reproduce memory, temporal inference, and computation. More broadly, they argued that what a nervous system can know is determined by the organization of the net itself. That ideas are not separate immaterial entities imposed on the brain, but patterns made possible by nervous organization. Although the model was biologically oversimplified, it helped lay the foundation for modern cognitive science, artificial intelligence, automata theory, and computational neuroscience.