Pavel Izmailov

Pavel Izmailov is a computer scientist and machine learning researcher at the Courant Institute of Mathematical Sciences at New York University. He is widely known for foundational work on neural network optimization and generalization, including the development of Stochastic Weight Averaging (SWA) and Bayesian deep learning methodologies for large-scale architectures.

Alongside Andy Q. Han and David Chalmers, Izmailov co-authored the 2026 paper “How’s it going? Reinforcement learning in language models recruits a functional welfare axis” (arXiv:2605.30232). The study uses geometric interpretability techniques to demonstrate that reinforcement learning algorithms recruit pre-existing linear subspaces encoding reward and punishment vectors, formalizing how goal-directed training activates latent evaluative dimensions embedded within pretrained model representations.

Izmailov’s broader research bridges theoretical machine learning and representation engineering. By examining the geometric properties of high-dimensional parameter spaces and activation manifolds, his work provides the mathematical foundation for analyzing how complex artificial networks represent task utility, uncertainty, and internal functional states.

Known for. Stochastic Weight Averaging (SWA), Bayesian deep learning, loss surface geometry, representation engineering in language models

Coverage on this site

• David Chalmers and Andy Han Find Reinforcement Learning Recruits a Functional Welfare Axis in Language Models September 2026

Related researchers

• David ChalmersNew York University. Co-director, Center for Mind, Brain and ConsciousnessNaming the hard problem of consciousness, and the philosophical zombie argument Functionalist16 articles
• AH Andy Q. HanCenter for Mind, Brain and Consciousness, New York UniversityMechanistic interpretability of welfare representations, concept steering in language models, functional welfare axes in reinforcement learning Functionalist1 article
• HA Hedayat AbedijooACE Conscious Studio / IndependentACE.await (2026), the ACE decision framework (Agency, Connection, Exchange) Functionalist1 article
• Blaise Agüera y ArcasGoogle VP and Fellow, CTO of Technology & SocietyComputational functionalism, active inference, on-device neural computing, What Is Intelligence? (2025), and Who Are We Now? (2023) Functionalist1 article
• IA Igor AleksanderImperial College LondonAxiomatic Consciousness Theory, neural state machines, The World in My Mind, pioneering neural network engineering Functionalist
• Joscha BachCalifornia Institute for Machine ConsciousnessThe machine consciousness hypothesis, cyber animism, and MicroPsi Functionalist8 articles

Pavel Izmailov is listed with the other functionalists and computationalists, who hold that what matters is the organisation of the processing, not the material. The theory family behind that position is set out in the index of consciousness theories. Every researcher covered on this site is indexed in the directory of consciousness researchers.