hiyenwong/ai_collection - Página 76

llm-serving-system-adaptive-architecture

LLM服务系统自适应架构设计 - 自进化系统、解耦架构、冷启动优化、黑盒调度、能效优化的综合技能框架。激活词: llm serving, adaptive architecture, autopoiesis, lora serving, llm cold start, inference scheduling.

Idioma do texto original: chinês

llm-sysml-alignment

LLM-assisted semantic alignment methodology for SysML v2 model integration in collaborative MBSE. Use when working with cross-organizational system model integration, SysML v2 semantic alignment, or LLM-based MBSE workflows. Keywords: SysML, MBSE, LLM,…

Idioma do texto original: inglês

local-gradient-approximations-rnn

Dynamics and Representation Structure of Local Approximations to Gradient-Based Learning in Linear Recurrent Neural Networks. Analytical framework comparing RFLO, tBPTT, and BPTT learning dynamics using dynamical systems theory. Key finding: RFLO solutions…

Idioma do texto original: inglês

logact-agentic-reliability

LogAct - enabling agentic reliability via shared logs. Deconstructed state machine architecture for LLM agents with pre-execution validation, failure recovery, and semantic introspection. Activation: agent reliability, agentic system, shared log, agent…

Idioma do texto original: inglês

low-rank-rnn-learning-dynamics

Framework for analyzing learning dynamics in low-rank RNNs via overlap space decomposition. Distinguishes loss-visible overlaps (determine activity/output/loss) from loss-invisible overlaps (encode training history). Enables understanding of why functionally…

Idioma do texto original: inglês

maximum-entropy-connectivity-networks

Maximum entropy principle for neural network connectivity — describe connectivity as a probability distribution over single-neuron weights, express task requirements as constraints, maximize Shannon entropy. From arXiv:2605.25607.

Idioma do texto original: inglês

maximum-entropy-network-structure-function

Maximum entropy principle for neural network connectivity that reveals how task constraints shape neural population structure without dependence on training procedure. Use when analyzing neural connectivity patterns, studying structure-function relationships,…

Idioma do texto original: inglês

memory-centric-agentic-research

Memory-centric agentic system methodology for full scientific research lifecycle automation. Covers schema-governed research memory (SciMem), five-stage lifecycle execution (SciFlow), DAG-shaped multi-agent operators (SciDAG), and self-evolving feedback loops…

Idioma do texto original: inglês

memory-uncertainty-relation-recurrent-networks

Memory Uncertainty Relation in random recurrent networks: inequality bounding short-term memory from below as an uncertainty relation between memory capacity and state-space fluctuations. Defines harmonic memory as an analytically tractable lower bound…

Idioma do texto original: inglês

meta-learning-biological-plasticity

Meta-Learning Biologically Plausible Plasticity Rules

Idioma do texto original: inglês

meta-learning-in-context-decoding-v3

Meta-learning In-Context Enables Training-Free Cross Subject Brain Decoding. Uses meta-learned in-context learning to decode brain signals across subjects without any subject-specific training. Supports visual decoding from fMRI/EEG signals with zero-shot…

Idioma do texto original: inglês

mistake-gated-continual-learning

Mistake-gated learning for energy and memory efficient continual learning using neuromorphic hardware. Only neurons that "make mistakes" (prediction errors) are updated, reducing compute and memory. Achieves 10-100x energy reduction vs full backprop on…

Idioma do texto original: inglês

momenta-multimodal-moe-misinformation-detection

MOMENTA — Mixture-of-Experts over multimodal embeddings with neural temporal aggregation for misinformation detection. Combines modality-specific MoE modules, bidirectional co-attention, discrepancy-aware branch, and attention-based temporal aggregation with…

Idioma do texto original: inglês

multi-agent-active-inference-digital-twins

Multi-agent digital twin framework using Active Inference for decentralized strategic decision-making. Features contextual inference and weighted message passing for coordination. Activation: active inference, multi-agent, digital twins, strategic…

Idioma do texto original: inglês

multi-agent-clinical-reasoning

Multi-agent framework for clinical reasoning and radiology AI. Use when designing multi-agent systems for medical diagnosis, radiology report generation, clinical decision support, or multi-modal medical reasoning. Triggers: multi-agent radiology, clinical…

Idioma do texto original: inglês

multi-agent-density-control

Stochastic Density-Driven Optimal Control (D²OC) for multi-agent coverage and distribution matching. Uses Wasserstein distance as running cost with convergence guarantees for stochastic LTI systems. Use when designing decentralized multi-agent coverage, area…

Idioma do texto original: inglês

multisensory-learning-engram-recruitment

Multisensory learning methodology that recruits visual neurons into olfactory memory engrams through cross-modal binding. Using Drosophila model to study how combining sensory modalities expands memory engrams and improves recall performance. Activation…

Idioma do texto original: inglês

muon-ogd-spectral-orthogonal-gradient-projection

Muon-OGD: Spectral-norm-aware orthogonal gradient projection for LLM continual learning. Integrates Muon optimizer's spectral-norm geometry with OGD's non-interference constraints. Activation triggers: Muon-OGD, spectral norm continual learning, orthogonal…

Idioma do texto original: inglês

near-policy-distillation

Methodology for accelerating on-policy distillation via asynchronous generation and evaluation. Decouples generation from evaluation using a near-policy buffer, achieving 2-4x throughput gains without quality degradation.

Idioma do texto original: inglês

nerve-network-aware-bilinear-fc-tokenization

NERVE: Network-Aware Representations of Brain Functional Connectivity via Bilinear Tokenization. Self-supervised learning framework for FC representation using network-aware bilinear tokenization in MAE. Partitions FC matrices into intra/inter-network…

Idioma do texto original: inglês

network-attractors-delay-plasticity

Network Attractors driven by Time-Delay Plasticity — framework for collective frequency selection and attractor formation via adaptive axonal delays (AADs), motivated by activity-dependent myelination in the brain. Uses delay-coupled phase oscillators on…

Idioma do texto original: inglês

network-aware-iv-regression

Network-aware Instrumental Variable Regression for Causal Node Discovery and Estimation. Two-stage framework incorporating IVs and graph-fused regularization for sparse causal effects in network-structured exposures with latent confounding. Activation:…

Idioma do texto original: chinês

neuro-bursty-persistent-networks

Bursty Persistent Brain Network (PBN) modeling methodology for neural dynamics with non-Markovian temporal structure. Combines renewal theory, state-dependent intensity functions, and stochastic simulations to model how neuronal avalanches transition between…

Idioma do texto original: inglês

neuro-dendritic-balance-learning

Dendritic balance learning methodology for predictive processing in cortical circuits. Combines compartmental neuron models with predictive coding principles, using dendritic prediction errors to drive synaptic plasticity. Applies to spiking neural networks,…

Idioma do texto original: inglês

neuro-sparse-deconvolved-predictive-network

Sparse Deconvolved Predictive Network methodology for neural dynamics modeling. Combines sparse coding, deconvolution of hemodynamic/synaptic responses, and predictive temporal modeling for extracting neural dynamics from observed signals. Applies to…

Idioma do texto original: inglês

neurocybernetic-modeling-large-scale

Integrative neurocybernetic modeling in the era of large-scale neuroscience. Closed-loop brain-body-environment models, nonlinear state-space, meta-dynamical extensions, knowledge distillation, connectomics-informed architectures. Trigger words:…

Idioma do texto original: inglês

nonlinear-rnn-fixed-connectivity-solution

Analytical solution for large nonlinear recurrent neural networks at fixed connectivity. Calculates moments and response functions without synaptic weight averaging, linking connectivity to spontaneous activity and perturbation response. Trigger words:…

Idioma do texto original: inglês

nonlinear-rnn-linear-equivalence

Linear equivalence of nonlinear recurrent neural networks using two-site cavity method. Shows covariance matrix of large nonlinear RNNs takes same form as linear networks with mean-field order parameters. Activation: nonlinear RNN, linear equivalence, cavity…

Idioma do texto original: chinês

noracl-neurogenesis-continual-learning

NORACL: Neurogenesis for Oracle-free Resource-Adaptive Continual Learning. Uses biologically-inspired neuronal growth to address the stability-plasticity dilemma without requiring oracle-sized architectures. Triggers: neurogenesis continual learning, adaptive…

Idioma do texto original: inglês

normalizing-trajectory-models

Normalizing Trajectory Models (NTM) methodology for few-step generative modeling with exact likelihood. Combines shallow invertible blocks within each denoising step with a deep parallel trajectory predictor, enabling end-to-end training and self-distillation…

Idioma do texto original: inglês

pbkv-agent-workflow

Prediction-based KV-Cache management for efficient serving of dynamic agent workflows. Predicts future agent invocations to optimize cache eviction and prefetching.

Idioma do texto original: inglês

physical-foundation-models-pfm

Physical Foundation Models (PFMs) — Fixed hardware implementations of large-scale neural networks where parameters are realized directly in physical substrate dynamics. Use when: designing specialized inference hardware, exploring optical/nanoelectronic…

Idioma do texto original: inglês

physical-foundation-models

Physical Foundation Models (PFMs): Fixed hardware implementations of large-scale neural networks realized directly in physical materials. Covers optical, nanoelectronic, and other physical platforms for trillion-parameter models. Activation: physical neural…

Idioma do texto original: inglês

pinns-biomedical-modeling

Physics-informed Neural Networks (PINNs) for biomedical modeling and simulation. Use when working on physics-guided neural network approaches for hemodynamics, cardiovascular modeling, blood flow prediction, or inverse medical physics problems. Combines…

Idioma do texto original: inglês

plant-model-mismatch-mpc

Model Predictive Control under plant-model mismatch - stability and suboptimality guarantees. Handles model uncertainty in control systems. Activation: MPC, model mismatch, robust control, plant-model mismatch, uncertainty in control systems.

Idioma do texto original: chinês

plasticity-network-framework

Network-based operationalization of plasticity as the ratio between system size and connectivity strength. Links structure to dynamical regimes (plastic vs rigid). Use for: complex systems analysis, brain plasticity quantification, neural network rigidity,…

Idioma do texto original: inglês

plasticity-prediction-deep-continual-learning

Theoretical framework for predicting plasticity in deep continual learning — understanding why neural networks lose their ability to adapt after training on previous tasks (loss of plasticity). Activation triggers: loss of plasticity, plasticity prediction,…

Idioma do texto original: inglês

poisson-gradient-estimation

Systematic comparison of Poisson gradient estimation methods (EAT vs GSM) for latent variable models in computational neuroscience. Activation: poisson gradient, EAT method, Gumbel-SoftMax, spike train inference.

Idioma do texto original: inglês

polystep-gradient-free-training

Gradient-free neural network training via Optimal Transport geometry (PolyStep optimizer). Based on arXiv:2605.01928 (Le, 2026). Use when training non-differentiable models including hard-LIF spiking neurons, quantized networks, discrete routing, or blackbox…

Idioma do texto original: inglês

polystep-optimal-transport-training

Gradient-free optimization for non-differentiable networks using optimal transport (PolyStep). Trains spiking neurons, quantized layers, discrete routing without surrogate gradients. Activation: polystep, optimal transport training, gradient-free optimizer,…

Idioma do texto original: inglês