research-code

Training-free anti-recomputation for video vision-language models.

Python/PySide research tool for visualizing community evolution in time-varying ECoG brain networks

Code for DyLoc: Dynamic Localization for Massive MIMO Using Predictive Recurrent Neural Networks (IEEE INFOCOM 2021).

🔁 REPRODUCE — Turn any prior empirical paper into a clean, minimal, runnable implementation. Compare, extend, reproduce.

DSP-inspired multirate wrappers for GPT with adaptive hyperparameters and faster training.

Non-LLM AI substrate that remembers who it is across sleep gaps. Pure numpy, Hebbian + sleep replay + persistent slow weights. Ten identity tests passing in toy worlds. Research artifact.

Autonomous Red Team: Multi-Agent Adversarial Security Testing — paper + proof-of-concept framework | DAI-2513 | Dissensus AI Working Paper

Research code and reproducibility materials for coarse- and fine-grained time-series forecasting using Gaussian process blurring and denoising.

Simulated Scholar Search (S3): scientific-literature environment and verl-based RL research code for search agents.

Scripts to interpret Mixture-of-Experts (MoE) language models at the expert level

Robust opponent modeling via adversarial ensemble training — core algorithm built on evolutionary RL, plus a modified multi-particle game environment.

Adversarial robustness evaluation framework for ML-based NIDS. Categorical robustness framework, the False Champion Problem. Backs IEEE Access submission. DoD-funded research with West Point.

Constraint-enforcing synthetic IoT packet generation. Two methods: statistical learning (PCA + dual OCSVM/IF gating, ~1,091 pkts/sec) and a genetic algorithm (composite fitness, ~5.7 pkts/sec, 0.62% anomaly). Amplifies the 5-sample ARP Spoofing class by 200x. All 12 ACI-IoT-2023 categories pass independent validators. ICCCN 2026 submission.

Minimal Tensor Logic AI engine in PyTorch – Datalog reasoning as tensor operations (einsum) and neural networks in one framework.

ASRI: Aggregated Systemic Risk Index for Cryptocurrency Markets — paper + real-time monitoring system | DAI-2509 | Dissensus AI Working Paper

TapeLM: facts as fingerprints on character ink — not token-id memory, not chunk RAG. One frozen curve encoder for generation and structured slot memory (write, bind, hop, resolve). Noisy recall, lexicon calibration, one-shot edits, clean unlearning — vs fair GPT/RAG; reproducible JSON benchmarks.

Hybrid LSTM-Markov attack chain forecasting for MITRE ATT&CK. Learns from 4,849 campaign chains + 8,437 real intrusion traces. Generates 26,051 risk-ranked multi-step attack futures via constrained beam search. 86% next-step accuracy, 0.76 Pearson correlation with NCISS severity. SECRYPT 2026 submission.

Official implementation of the Lattice Stabilization Protocol for AGI Alignment (Type-IV). Implements zero-knowledge tensor bridging for hyper-dimensional manifolds.