ablation

Ablation is an autonomous reverse engineering system

Model Ablation Tool-Kit for Deep Learning Model

Distribution transparent Machine Learning experiments on Apache Spark

The production engine for directional ablation. Unalign / remove models censorship efficiently on any hardware.

Extract and evaluate radiomics for liver cancer tumors from DICOM segmentation masks. Using SimpleITK, PyRadiomics and PyDicom.

Explainability of Deep Learning Models

Evaluating XAI methods through ablation studies.

Explainability of Brain Tumour Segmentation Models

O(N) attention with a bounded inference KV cache. D4 Daubechies wavelet field + content-gated Q·K gather at dyadic offsets.

Library to compute 3D surface-distances for evaluating liver ablation/tumor completeness based on segmentation images.

Clean-room behavioral model editing and ablation engine with capability-preserving PRIME optimization.

Multi-agent verification for AI outputs: claim verification, RAG diagnostics, pre-action verification for agentic AI. Includes ablation studies proving multi-agent vs single-prompt tradeoffs, FaithBench benchmarks, and bias-triggering evaluation methodology

Does a CLAUDE.md actually change how Claude behaves? An ablation harness: run adversarial traps with the rules and without them, grade blind, and test whether the difference is real.

A layered prompt engine with ablation-based evaluation for long-form fiction. 五层信息阶梯 · 逐层充分性消融 · 双向阶梯 · AI 味防护 · PromptOpt 两层共演化。改编层 / 资产管线 / 漫剧线 / ark 控制台 / Electron 桌面。

Causal Chain-of-Thought step faithfulness evaluation harness — quantifies whether reasoning steps actually drive model outputs or are decorative

Controlled ablation bench for tool-using agents: tools, skills, and model choice with bootstrap CIs.

Genetics for Algorithm Discovery. Evolve ideas, not programs: every LLM idea becomes a named switch, and knockouts measure what each is worth. New lower bounds at 17 sizes of Tao et al.'s problem 60. Built at the London AI x Science Hackathon 2026.

CIFAR-10 CNN: 0.8003 -> 0.9020, one variable at a time, 3 seeds per config; every table generated from the raw run logs.

Measure which agents actually influence multi-agent decisions using leave-one-out ablation.

External memory infrastructure for AI agents — measurement frameworks, multi-agent orchestration, metacognitive self-improvement. 9 independent convergences. Open methodology + implementation scripts.