Neural Programmer-Interpreter Implementation (Reed, de Freitas: https://arxiv.org/abs/1511.06279), in Tensorflow
-
Updated
Nov 17, 2018 - Python
Neural Programmer-Interpreter Implementation (Reed, de Freitas: https://arxiv.org/abs/1511.06279), in Tensorflow
A Word Level Transformer layer based on PyTorch and 🤗 Transformers.
CPU activation probes for AI-agent guardrails and classification. One frozen local model pass, many independent heads. Open models, code and reproducible benchmarks.
IJRR 2026 | Situationally-Aware Dynamics Learning | Online and Unsupervised Latent Factor Representation Learning for Robot Dynamics Learning.
Visual analytics approach presented in the paper "Visual Analytics Tool for the Interpretation of Hidden States in Recurrent Neural Networks" (VCIBA, 2021).
Pre-generation tool-call gating via linear probes on LLM hidden states. F1 ≈ 0.91–0.94 on BFCL v4, 14–22× faster than full generation. Cross-architecture transfer across Llama / Qwen / Phi / Mistral (3B–7B) with ≥96% retention.
A J-space-inspired AI visual art and interpretability playground for watching hidden-state word candidates swarm and collapse into language.
Designing and training probabilistic graphical models (MATLAB).
R package for Statistical Modeling of Animal Movements
This repository contains NLP Transfer learning projects with deployment and integration with UI.
Layer-wise hidden-state probing for early detection of harmful intent in small instruction-tuned language models.
Compare layer-wise hidden states from aligned twin prompts, reporting mean token cosine similarity, L2 distance, first threshold-based divergence, and largest-change layer. Works with (T, D) states from any framework.
Coordinate navigation for Transformer internal representations — zero training, any model
A Python tool for training seeded logistic probes on model hidden states and comparing per-layer train accuracy to estimate where a property is linearly readable.
Python package for measuring hidden-state consensus across paraphrases, reporting layer-wise mean and minimum pairwise cosine agreement, the first-disagreement layer, consensus peak, and final agreement. Measures representational similarity, not answer correctness.
A Python toolkit for measuring per-layer model hidden-state geometry: covariance-based intrinsic dimension, anisotropy (mean pairwise cosine), and first-PC variance share, with a layer-by-layer report.
Investigating Layer-Specific Performance in Speaker Recognition with XLS-R Architecture
Probes of the gap between recoverable hidden-state semantics and decoder outputs
Code, datasets and LaTeX source of my MSc thesis: detecting routing failures in LLM-based multi-agent systems with linear probes on hidden states
Principal component geometry and feature readout in Transformer hidden states (GPT-2, OPT, Qwen2.5). Shows 96–98% PC1 variance concentration is driven by the first token, while PC2–PC8 preserve transferable token-level surprisal across articles.
To associate your repository with the hidden-states topic, visit your repo's landing page and select "manage topics."