net-inspection-cv Public

Honest CV prototype for aquaculture fish-farm net-damage detection. Compares 4 methods (classical OpenCV, hand-crafted + PatchCore anomaly, YOLOv8/seg) with adversarial evaluation, temporal reasoni...

opencv computer-vision deep-learning pytorch ros object-detection underwater

Python Other Updated Aug 20, 2026

secret-loyalty-probe Public

A calibration standard for weight-space secret-loyalty detectors: a min-max audit's false-positive floor is 2/(m+1), so claiming 5% costs 39 negative controls. Includes the audit whose own survivin...

statistics reproducible-research alignment lora ai-safety interpretability red-teaming

Python MIT License Updated Aug 18, 2026

Chrislysen Public

3 Updated Jul 23, 2026

deploy-doctor Public

Does your int8 'GPU' model actually run on the GPU? A PyTorch CLI that catches silent device-placement footguns (int8 CPU-lock, meta/offload/fp16/fp64) before production.

cli machine-learning deployment gpu pytorch developer-tools quantization

Python MIT License Updated Jun 6, 2026

opsem Public

Training-free, CPU-only lexical–dense fusion for conversational-memory retrieval — 0.752 vs 0.640 Hit@1 over BM25 on LoCoMo, fully reproducible. Paper + reproduction code.

nlp information-retrieval reproducibility bm25 conversational-ai rag dense-retrieval

Python 2 Other Updated Jun 2, 2026

SLO-Guard Public

Crash-aware autotuner for vLLM serving under hard latency/memory SLOs via feasibility-boundary learning — 150-trial A100 study: 75/75 feasible, ×ばつ tighter variance. Paper + datasets.

slo autotuning bayesian-optimization tpe mlops inference-optimization llm-serving

Python 1 Updated Apr 19, 2026

Constrained-ML-Deployment Public

Crash-aware two-phase optimizer (TBA→TPE) for constrained ML-deployment config search — 80% optimal-model discovery vs 30% for TPE across 5 NVIDIA GPUs. Two papers + the DeployBench suite.

benchmark gpu quantization bayesian-optimization automl onnx mlops

Python MIT License Updated Apr 10, 2026

qaoa-benchmark Public

Benchmarking budget-aware classical optimizers vs COBYLA/SPSA for QAOA parameter tuning on IBM quantum hardware — negative result: shallow QAOA landscapes are too smooth to benefit from constrained...

benchmark optimization quantum-computing qiskit qaoa negative-results quantum-optimization

Python Updated Apr 9, 2026

dwave-benchmark Public

Classical SA vs D-Wave quantum annealing on Max-Cut and spin glass problems. Phase 1 complete: all classical solvers converge to identical solutions up to n=500. Waiting on D-Wave QPU access to tes...

benchmark optimization simulated-annealing d-wave spin-glass quantum-annealing annealing

Python Updated Apr 9, 2026

Norwegian-AI-Championship Public

NM i AI 2026 entry (Team INNBerkeley) — 0.8807 mAP@0.5 grocery-shelf object detection (YOLOv8x + multi-scale TTA + WBF/classifier re-ranking, ONNX), a Gemini-powered FastAPI accounting agent (Tripl...

competition computer-vision ensemble object-detection onnx fastapi test-time-augmentation

Python Updated Apr 9, 2026

deploy-agent Public

Feasible-first exploration optimizer for constrained ML deployment in crash-prone hierarchical search spaces. TBA→TPE hybrid with DeployBench benchmark suite.

cli deployment optimization mcp mlops inference-optimization fastapi

Python Updated Mar 30, 2026

ML2-Exam Public

Machine Learning 2 exam — a TensorFlow/Keras CNN image classifier and a PyTorch convolutional variational autoencoder (VAE) generating ×ばつ64 CelebA faces.

deep-learning tensorflow keras cnn pytorch vae image-classification

Python Updated Dec 10, 2025