Data & Analytics

Transformer 表征轨迹几何分析方法论 - 将计算神经科学的几何工具应用于 Transformer 可解释性研究，无需探测即可分析表征动力学

IQP Quantum Circuit Born Machines trainability analysis under Gaussian initialization. Uses Stein's lemma and Lipschitz concentration bounds to derive analytical lower bounds on gradient variance and probabilistic concentration bounds for barren plateau avoidance in QCBMs. Activation: IQP circuit, Born machine, QCBM trainability, barren plateau, gradient concentration, Gaussian initialization, quantum generative model, MMD loss, Stein's lemma, Lipschitz bound, quantum machine learning

Topological Machine Learning for epileptic iEEG seizure detection using persistent homology and persistence diagrams. Features multiple TDA representations and cross-patient generalization. Activation: topological data analysis, TDA, EEG classification, seizure detection, persistent homology.

TMS-EEG生物标志物信效度评估方法论。系统评估TMS-EEG标志物的内部可靠性、外部可靠性和有效性，提供评估框架和最佳实践。触发词：TMS-EEG、生物标志物、可靠性、有效性、信效度、TMS biomarkers、reliability、validity、TMS-EEG analysis。

Non-equilibrium thermodynamic framework for quantum reservoir computing - links predictive performance to energetic costs via Holevo capacities and quantum informational dissipation

Thermodynamic framework for analyzing multiplex neural connectomes, linking synaptic and neuropeptidergic signaling layers. Applied to the complete C. elegans connectome to reveal functional specialization and hierarchical organization through energy-based connectivity analysis.

**arXiv ID:** 2405.07987 **Authors:** Minyoung Huh, Brian Cheung, Tongzhou Wang, Phillip Isola **Published:** 2024-05-13T17:58:30Z **Abstract:** We argue that representations in AI models, particularly deep networks, are converging. First, we survey many examples of convergence in the literature: over time and across multiple domains, the ways by which different neural networks represent data are becoming more aligned. Next, we demonstrate convergence across data modalities: as vision models ...

Tensor network diagram methodology for simplifying tensor algebra - graphical notation for contractions, decompositions, and gradient computation bridging quantum physics notation with machine learning. Activation: tensor network diagrams, tensor cookbook, penrose notation, tensor contraction diagrams, 张量网络图, 张量图解.

**arXiv ID:** 2105.13336 **Authors:** Kaixin Zhang, Hongzhi Wang, Han Hu, Songling Zou, Jiye Qiu, Tongxin Li, Zhishun Wang **Published:** 2021-05-27T17:46:16Z **Abstract:** Recently, deep learning has been an area of intense research. However, as a kind of computing-intensive task, deep learning highly relies on the scale of GPU memory, which is usually prohibitive and scarce. Although some extensive works have been proposed for dynamic GPU memory management, they are hard to apply to systems...

Brain Neural Operator (Tau-BNO) surrogate framework for rapidly approximating Network Transport Model dynamics of pathological tau protein spread in Alzheimer's disease. Combines function operator encoding kinetic parameters with query operator preserving initial state, using spectral kernel for anisotropic transport. Activation triggers: tau propagation, alzheimer modeling, neural operator, brain network transport, biophysical surrogate, disease progression modeling.

Beyond Prediction Accuracy: Target-Space Recovery Profiles for Evaluating Model-Brain Alignment — a framework for identifying which reproducible brain response dimensions are recovered by model predictions, going beyond simple prediction accuracy. (arXiv:2605.20127)

Spectral analysis of synaptic matrix eigenvalues for stability, transient dynamics, and memory capacity analysis in sparsely connected neural networks

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 learning to support computation of covariance structure in neural networks.

**arXiv ID:** 2203.09250 **Authors:** Irina Higgins, Sébastien Racanière, Danilo Rezende **Published:** 2022-03-17T11:18:34Z **Abstract:** Biological intelligence is remarkable in its ability to produce complex behaviour in many diverse situations through data efficient, generalisable and transferable skill acquisition. It is believed that learning "good" sensory representations is important for enabling this, however there is little agreement as to what a good representation should look like...

First systematic application of Successor Representations (SRs) from reinforcement learning to natural language. Trains deep residual network on WikiText-103 to predict future word distributions; structured language representations (noun/verb/adjective categories) emerge spontaneously without explicit linguistic supervision. Establishes bridge between RL, linguistics, and cognitive neuroscience. Based on arXiv:2605.24585 (May 2026). Use when studying successor representations in language, eme...

Subcortical shape variations and their associations with cognition across the 8th decade of life. Longitudinal study using neuroimaging and cognitive data from Lothian Birth Cohort 1936. Analyzes heterogeneous morphological trajectories in hippocampus, thalami, globus pallidi, and ventral DC. Uses ANCOVA and mixed linear model analyses to investigate vertex displacement patterns associated with cognitive aging. Use when studying brain morphology changes, subcortical shape analysis, cognitive ...

Early preconfiguration failure detection methodology for repetitive subconcussive (rSC) brain injuries using high-density EEG. Captures millisecond-level cortical dynamics and spatiotemporal features for sports neurology and concussion screening. Activation: subconcussion, EEG, sports neurology, concussion detection, brain injury.

Structure-aware variance reduction methodology for unbiased randomized Hamiltonian simulation. Combines classical variance reduction with randomized product-formula estimators to achieve 70-96% sampling cost reductions in tensor-network simulations. Use when implementing randomized Hamiltonian simulation, optimizing quantum circuit sampling, reducing Trotter discretization errors, or analyzing non-commutative Hamiltonian dynamics.

Structure-Activity in Nonlinear Spiking Networks

Analysis methodology for structural plasticity in neural networks — evaluating growth vs pruning operators, newborn unit integration stability, and time-sensitive optimization dynamics. Covers forward-active backward-starved phenomenon, insertion stability, and continual learning plasticity.

Comprehensive stock technical analysis system for fetching data, calculating indicators (KDJ, MACD, RSI, BOLL), generating visualizations and reports. Use when user asks about stock analysis, 股票分析, technical analysis, 技术分析, k-line, or stock scoring.

Stochastic Graph Heat Modelling methodology for brain connectivity estimation. Uses noise-driven heat diffusion on graphs to estimate directed, multivariate, dynamic, model-based connectivity from neurophysiological data. Extends traditional coherence methods with graph-based PDE formulation and regularization. Activation: brain connectivity, graph heat modelling, neurophysiological data, directed connectivity, coherence, graph PDE, effective connectivity

Stimulus symmetries can confound representational similarity analyses — demonstrates how stimulus symmetries in neural network inputs cause functionally-equivalent representations to produce different, drifting RSM geometries. Based on arXiv:2605.21324.

Extended STDP learning rule for simultaneously learning synaptic connection strengths and delays, validated on unsupervised SNN classification tasks with superior performance over delay-free STDP.