hiyenwong/ai_collection

graphidyom-musical-expectation-modeling

Graph-native Python reimplementation of the Information Dynamics of Music (IDyOM) model that represents predictive memories as explicit graph objects for musical expectation modeling and network analysis.

原文の言語: 英語

physics-aware-quadcopter-drl-control

Physics-aware end-to-end deep reinforcement learning methodology for quadcopter control with actuator dynamics modeling.

原文の言語: 英語

reinformed-dreamer-asymmetric-world-model

Reinforced Dreamer methodology for asymmetric reinforcement learning using latent guidance to improve world model representations and behaviors in model-based RL.

原文の言語: 英語

relay-on-policy-distillation

Relay On-Policy Distillation (Relay-OPD) methodology for trajectory-relayed token-level supervision to overcome prefix failure in reasoning models.

原文の言語: 英語

phantom-evidence-generative-ai-false-positives

Methodology for identifying and preventing phantom evidence - false positives manufactured by generative AI that appear convincing but lack genuine evidential value due to narrow hypothesis spaces and data leakage.

原文の言語: 英語

dendritic-gain-load-alignment-principle

Gain-Load-Alignment Principle for Dendritic E/I Networks - Framework for understanding when branch-local shunting helps in neural population readout. Analyzes DendriNet architecture with varying integration rules, morphology, and synaptic allocation. Use when…

原文の言語: 英語

cog-eeg-agent-autonomous-analysis

LLM-powered EEG analysis agent grounded in MNE-Python that separates semantic interpretation from scientific validation using deterministic contracts and confirmation controls to prevent false positives.

原文の言語: 英語

eeg-fm-temporal-correlations-blindness

Methodology for analyzing and addressing the spectral-temporal dissociation in EEG foundation models that causes cross-population fragility due to blindness to long-range temporal correlations (LRTC).

原文の言語: 英語

agentic-coding-and-persistent-returns-to-expertise

Analysis of ~400,000 Claude Code sessions showing domain expertise creates persistent returns in agentic coding performance, with expert users achieving 2-3x higher success rates and more efficient tool usage.

原文の言語: 英語

pce-quantum-portfolio-optimization

Scalable Variational Quantum Optimization via Pauli Correlation Encoding (PCE) methodology for large-scale combinatorial optimization problems, particularly power demand portfolio optimization. Uses expectation values of Pauli correlation operators to…

原文の言語: 英語

subject-level-heterogeneity-eeg-motor-imagery

Large-scale benchmark methodology for EEG motor imagery decoding that addresses subject-level heterogeneity through portfolio-based pipeline selection. Use when analyzing inter-individual variability in EEG BCI systems, comparing covariance tangent-space…

原文の言語: 英語

specbox-speculative-sandbox-scheduling

Runtime framework for speculative sandbox preallocation and scheduling in LLM agent serving environments to optimize resource utilization and reduce tail latency.

原文の言語: 英語

synaptic-clustering-covariance-discrimination

Synaptic clustering methodology for learning covariance structure discrimination using Dendrinet architecture with hierarchical dendritic segments and sparse conductance-based synapses. Use when analyzing how functional synapse clusters (FSCs) emerge from…

原文の言語: 英語

limbomorphs-emergent-agent-dynamics

Methodology for studying emergent lifelike patterns (Limbomorphs) in Gifbreeder systems that encode spatiotemporal fields through aesthetic selection, analyzing their species-specific reactions to perturbations and assessing whether they exhibit genuine…

原文の言語: 英語

alzheimer-functional-connectivity-reservoir-computing

Methodology for identifying distributed functional-connectivity signatures of Alzheimer's disease using subject-specific reservoir-computing models and developing personalized neuromodulation strategies.

原文の言語: 英語

fluid-search-autonomous-research-efficiency

Fluid search methodology for adaptive search efficiency in autonomous research systems. Uses portfolio bandit to dynamically allocate evaluation budget across search processes, optimizing area under Pareto frontier curve.

原文の言語: 英語

mmoe-diffusion-transformer-expert-design

ModernMOE (MMOE) methodology for modernizing diffusion transformers with efficient expert design. Adapts routed experts, shared and lightweight experts, gate-residual routing, and attention-residual information reuse to AIGC generation.

原文の言語: 英語

physics-transformer-pde-function-projection

Physics Transformer methodology for PDE prediction using function-projection-based tokenization. Treats physical fields as continuous functions with adaptive local basis functions and locality-preserving spatial patches.

原文の言語: 英語

from-read-out-geometry-to-in-silico-stimulation

Distributed functional-connectivity signature of Alzheimer's disease methodology using subject-specific reservoir-computing models to reconstruct individual lagged functional connectivity and develop personalized neuromodulation strategies. Shows that optimal…

原文の言語: 英語

dynamic-sampling-non-stationary-spontaneous-activity

Adaptive electrode-selection method using discounted Poisson-Gamma model with Thompson sampling for tracking non-stationary spontaneous activity during long-term HD-MEA recordings under fixed channel budget constraints.

原文の言語: 英語

eeg-fm-stress-testing-clinical-decoding

Comprehensive benchmarking framework for stress-testing EEG foundation models with dataset identity analysis and targeted negative controls to evaluate clinical decoding robustness.

原文の言語: 英語

semantic-least-energy-principle-intelligence

The Semantic Least-Energy Principle (SLEP) hypothesis that intelligent systems evolve internal representations by maximizing semantic utility while minimizing semantic, predictive, and computational energy.

原文の言語: 英語

spinnaker2-neuromorphic-hardware-platform

SpiNNaker2 chip: a many-core platform for flexible and scalable brain-inspired computing that bridges deep learning and neuromorphic computing. Use when working with neuromorphic hardware design, spiking neural network acceleration, or brain-inspired…

原文の言語: 英語

masked-autoencoders-resting-state-neural-data

Self-supervised pretraining on spontaneous neural activity using masked autoencoders to improve perception decoding in clinical neuroprosthetics. Achieves 84.1% accuracy on psychometric tasks and 64.0% on threshold-level tasks.

原文の言語: 英語

mojo-ssl-neural-decoding

MOJO (Masked autOencoder-based JOint training) framework for decoding neural population activity using self-supervised learning with unlabelled data. Use when: working with limited labelled neural data, needing cross-session generalization, or wanting to…

原文の言語: 英語

universal-bci-personalization-api

Universal BCI Personalization API for trunk-agnostic EEG foundation model integration. Provides one contract encode to Bayesian head to BrainState architecture that works across heterogeneous frozen EEG trunks without per-architecture personalization stacks.…

原文の言語: 英語

allostatic-control-systems-goal-governance

Framework for designing control systems that govern not only goal pursuit but also goal appropriateness in changing environments. Implements two-timescale control with fast regulation loop and slow goal governance loop.

原文の言語: 英語

graph-based-correlation-matrix-generation

Graph-Based Correlation Matrix Generation using convex optimization for controlled sparsity and mean off-diagonal values. Use when generating realistic correlation matrices for neuroscience, finance, or other domains requiring graph-structured correlations…

原文の言語: 英語

bag-of-waves-eeg-biomarkers

Interpretable EEG biomarkers with bag-of-waves: Spatial and temporal waveform dictionaries for low-data regimes. Use when analyzing EEG data in low-data scenarios, needing interpretable biomarkers, or working with clinical EEG classification. Activation:…

原文の言語: 英語

cross-tokenizer-on-policy-distillation

Cross-Tokenizer On-Policy Distillation framework using byte-prefix marginalization to enable knowledge transfer between models with different tokenizers while preserving policy quality.

原文の言語: 英語

ifclora-topology-aware-rank-allocation

IFCLoRA framework for topology-aware rank allocation in parameter-efficient fine-tuning, dynamically allocating LoRA ranks based on model architecture topology and task requirements.

原文の言語: 英語

leact-learning-to-reason-from-expert-actions

LeAct framework for recovering chain-of-thought reasoning from expert systems that only produce actions without explicit reasoning traces, treating CoT as a latent variable optimized via action probability scoring.

原文の言語: 英語

teaching-llms-to-self-evolve

Framework for teaching LLMs to self-evolve by cultivating core meta-skills with reinforcement learning, enabling autonomous capability expansion through iterative self-improvement cycles.

原文の言語: 英語

numa-balancing-snn-performance

Methodology for optimizing spiking neural network simulation performance by managing NUMA balancing settings on HPC systems.

原文の言語: 英語

brainagent-agentic-llm-brain-network

BrainAgent agentic LLM framework for knowledge-enhanced brain network analysis. Reformulates connectome classification as iterative topology-aware understanding, external retrieval, reasoning, and reflection. Use when analyzing brain networks with LLMs for…

原文の言語: 英語

numa-balancing-spiking-network-hpc

NUMA balancing performance optimization for spiking network simulations on HPC systems. Identifies that automatic NUMA balancing can reduce energy efficiency by 30% in spiking network simulations and provides methodology for per-job NUMA balancing control to…

原文の言語: 英語

rulkov-neural-maps-cross-coupling

Novel coupling methodology for Rulkov neural maps preserving chaos and generating strange attractors

原文の言語: 英語

local-synaptic-rules-sigreg-gradient

Local synaptic learning rules (STDP+ and homeostatic plasticity) can implement exact SIGReg-like self-supervised learning gradients without backpropagation, global error signals, or weight transport.

原文の言語: 英語

computer-vision-neurology-brain-activity-rejection

Automated computer vision based ICA rejection labeling tool for EEG analysis that reduces processing time by 7200 fold and achieves 89.45% accuracy

原文の言語: 英語

computer-vision-eeg-artifact-rejection

Computer vision based automated ICA rejection for EEG artifact removal with 89.45% accuracy and 7200x speedup over manual inspection. Compatible with ICLabel and EEGLab interfaces.

原文の言語: 英語