Results for “statistical-learning-theory”

AutoAugment: Learning Augmentation Strategies from Data

Training Compute-Optimal Large Language Models

Train or fine-tune sentence-transformers models for retrieval, similarity, clustering, classification, and reranking, with support for bi-encoders, cross-encoders, and sparse encoders.

Train and fine-tune transformer language models using TRL (Transformers Reinforcement Learning) with support for SFT, DPO, GRPO, KTO, RLOO, and reward model training via CLI commands.

Train, evaluate, export, and run inference for Masked Auto-Encoder (MAE) models for self-supervised pretraining and fine-tuning of visual representations.

Train, evaluate, export, and run inference for pose classification models using ST-GCN on skeleton keypoint sequences.

Fine-tune and align language models using reinforcement learning with TRL, including SFT, DPO, PPO, GRPO, and reward model training.

Scaling Data-Constrained Language Models

Curriculum Learning

Scaling Instruction-Finetuned Language Models

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

Anki spaced repetition learning system reference. Covers the science of SRS and SM-2 algorithm, card design principles, optimal deck settings, FSRS scheduler, study workflow, essential add-ons, custom templates, filtered decks, and AnkiConnect API.

Foundational theory for process-oriented concurrency through synchronous message-passing, applicable to multi-agent coordination and parallel decomposition

Multimodal Few-Shot Learning with Frozen Language Models

Breaks a topic into a sequence of adaptive learning articles, adjusting depth and pace based on user feedback from the previous lesson.

Continuous self-improvement through systematic logging, pattern detection, and behavioral updates. Use when: the owner corrects you, a task fails, you discover a better approach, or you notice a recurring pattern. Store raw learnings in .learnings/, update the specific skill or workflow that caused the issue when appropriate, and avoid vague promises to do better.

Fine-tune instruction-following LLMs with Unsloth's optimized SFTTrainer, covering dataset formatting, chat templates, training configuration, and thinking-model patterns.

Post-train LLMs with reinforcement learning using the slime framework, which integrates Megatron-LM for training and SGLang for rollout generation.

Design a spaced retrieval schedule for any topic list and timeline. Use when planning units, term sequences, or revision programmes.

Visual Instruction Tuning

Wrap a learning session in a plan → monitor → reflect cycle. Use at the start of any substantial study session to set goals, mid-session to check strategy, and at session end to consolidate what changed. Builds self-regulated learning as a habit.

Influence Functions in Deep Learning

Sigmoid Loss for Language Image Pre-Training

Structured threat modeling skill using the PASTA framework (Process for Attack Simulation and Threat Analysis) combined with Adam Shostack's 4-question framework. Use this skill whenever the user asks to do threat modeling, security analysis, map the attack surface, identify threats, or review an application for security risks — even if they don't mention PASTA or a specific framework by name. Core activities: Component Mapping (architecture + data flows), Critical Assessment (business impact prioritization), and Logic Flaw Identification (attacker mindset on business logic). Produces: component map diagram (Mermaid), data flow diagram (Mermaid), attack tree (Mermaid), STRIDE threat table, prioritized risk register, and an actionable mitigation plan. Invoke proactively for any security review, architecture review, or "what could go wrong?" session.

Explain and apply core ML/AI concepts — model types, training pipelines, evaluation metrics, and neural architectures.

Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.

Hard Negative Mixing for Contrastive Learning

Build and evaluate machine learning models using scikit-learn for classification, regression, clustering, dimensionality reduction, and preprocessing.

Matryoshka Representation Learning

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.

Recursive self-improving holon λ(ο,Κ,Σ).τ' for knowledge compounding and schema evolution. USE WHEN learning, improving, optimizing, assessing, reflecting, debugging, synthesizing, or refining—whether human, AI, or organizational. Triggers on /learn, /compound, /improve, /refine, /optimize, /assess, /reflect, "lessons learned", "best practices", "continuous improvement". Preserves Κ-monotonicity, η≥4, homoiconicity.

Automatically evaluates Claude Code sessions to extract reusable patterns and save them as learned skills.

Modify a classroom task with language scaffolds that preserve cognitive demand for EAL learners. Use when adapting existing tasks for students at different English proficiency levels.

Scaling Laws for Neural Language Models

Design a Concrete-Pictorial-Abstract learning sequence for a mathematical concept using manipulatives. Use when teaching maths through Singapore method or when students struggle with abstraction.