Neuromorphic computing abstract

Frederick Hayes

Investigating how intelligent systems learn, preserve, and adapt internal representations under uncertainty.

Hi, I am Fred. I am a data scientist and independent AI researcher working at the intersection of machine learning, probabilistic reasoning, computational neuroscience, and mechanistic AI. My current research asks how adaptive agents preserve useful internal representations when the world becomes uncertain, information is degraded, or the agent's own latent state is at risk.

I recently completed the Master of Information and Data Science at UC Berkeley. Alongside applied machine learning work in biotech and industrial AI, I am developing Erasure Horizon, a research program on latent-state continuity, reward abandonment, and uncertainty-aware control in recurrent agents.

Adaptive Intelligent Systems Under Uncertainty

My research is organized around a simple question: what should an intelligent system do when the uncertainty is not only about the external world, but about the reliability and continuity of its own internal state?

That question connects my interests in recurrent neural networks, spiking neural networks, mechanistic interpretability, probabilistic inference, Markov models, and post-von Neumann AI architectures. I am especially interested in agents that can distinguish temporary sensory interruption from deeper internal vulnerability, and that can change control strategy when ordinary reward pursuit becomes dangerous.

In applied work, this same orientation shows up as rigorous evaluation: building models that are useful under messy real-world constraints, detecting failure modes, designing trustworthy benchmarks, and communicating uncertainty clearly to technical and non-technical stakeholders.

Research & Projects

A selection of recent work spanning adaptive agents, representation analysis, AI evaluation, and biologically inspired computation.

Erasure Horizon: Latent-State Continuity in Recurrent Agents

Current independent research program. Erasure Horizon studies recurrent agents that must distinguish ordinary I/O suspension from destructive internal-state threat. The project operationalizes a functional analogue of latent-state vulnerability: agents learn when reward pursuit should be subordinated to preserving the internal state that makes future action possible.

The single-agent phase combines procedural environments, GRU policies, reward-distractor calibration, hidden-state geometry, linear probes, actor-sensitive hazard directions, causal rollout interventions, temporal modulation experiments, and off-target activation checks.

Current supported findings:

  • Agents distinguish temporary sensory interruption from Erasure-like internal-state threat.
  • Erasure pressure produces reward abandonment or reward subordination.
  • Erasure warning produces a distinct hidden-state regime shift.
  • Actor-sensitive components of the hazard representation causally reduce Erasure exposure.
  • Temporal modulation of the hazard manifold improves Erasure-warning control without collapsing ordinary I/O bridging.

Manuscript and public code repository in preparation.

Internal Control Regimes for Behavior Representation in Partially Observable Agents

Single-author manuscript under review. This work investigates whether internal control regimes provide a richer representation of behavior than external action traces alone in partially observable agents. The study combines recurrent and spiking agents, internal-state probes, perturbation analysis, feature-family controls, and event-level prediction.

Key Focus Areas:

  • Behavior representation under partial observability
  • Recurrent policies and spiking neural networks
  • Internal-state analysis and perturbation methods

Submitted to SBP-BRiMS 2026.

NeuroBeacon: Calibrated RL for Cognitive Augmentation

NeuroBeacon Project

NeuroBeacon is a UC Berkeley MIDS capstone project selected for the Spring 2025 Capstone Showcase. The project explored reinforcement learning for cognitive augmentation in human-in-the-loop systems, emphasizing calibration, uncertainty signaling, game design, and transparent interaction rather than raw accuracy alone.

Key Focus Areas:

  • Deep Reinforcement Learning and adaptive difficulty
  • Calibration and uncertainty-aware decision support
  • Human-centered AI and cognitive augmentation

View the UC Berkeley Capstone Showcase Page

Audience-Adaptive AI Agents (RAG & LangGraph)

RAG System

Built a solo Retrieval-Augmented Generation system for Berkeley's Generative AI course using LangGraph for decomposition, retrieval, reranking, and audience-specific response generation. To evaluate whether outputs matched the intended audience and source evidence, I developed the Fred Score, a composite evaluation framework balancing retrieval alignment, semantic match, and response quality.

Key Focus Areas:

  • Agentic orchestration and multi-step reasoning
  • Human-centered evaluation and hallucination analysis
  • Audience-aware prompting and retrieval calibration

Read the Final Paper

Synthetic Dasein: Neuromodulated Plasticity

Explored the metabolic cost of cognition in dynamic GridWorld environments. Using phase-dependent gating of Spike-Timing-Dependent Plasticity (STDP), the project handed over agent control to a Spiking Neural Network and investigated how local learning, oscillatory coordination, and energy constraints shape adaptive behavior.

Key Focus Areas:

  • Spiking Neural Networks and continuous-time architectures
  • STDP, local plasticity, and neuromodulated learning
  • Oscillatory coordination and energy-aware computation

View the Repository on GitHub

Writing & Publications

I write about AI mechanism, scientific reasoning, and the philosophy of intelligent systems in my Medium column, Let's Talk About the Work.

Technical Toolkit & Research Methods

  • Research Methods: mechanistic interpretability, hidden-state probes, representation geometry, causal rollout interventions, perturbation studies, benchmark design, and multi-seed confirmatory experiments.
  • Machine Learning & Statistics: PyTorch, scikit-learn, XGBoost/CatBoost-style workflows, statistical modeling, Bayesian reasoning, Markov chains, MCMC interests, calibration, and uncertainty-aware evaluation.
  • Agents & Generative AI: LangGraph, LangChain, retrieval-augmented generation, LLM evaluation, human-in-the-loop systems, audience-aware prompting, and agentic workflow design.
  • Neuromorphic AI: Spiking Neural Networks, STDP, oscillatory coordination, event-driven computation, post-von Neumann architectures, and energy-efficient AI.
  • Languages & Engineering: Python, SQL, R, modular software design, Git, Docker, cloud-based ML workflows, Azure infrastructure, and reproducible experiment pipelines.

Contact Me

I am always glad to discuss research ideas, collaborations, PhD opportunities, applied AI problems, or thoughtful conversations about adaptive intelligent systems.

You can reach me by email or connect with me via LinkedIn.

My Curriculum Vitae

You can view or download a copy of my research CV to learn more about my education, research interests, selected projects, and professional experience.