Data & Analytics

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 method, mean-field analysis, covariance matrix.

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: nonlinear RNN, fixed connectivity, moments, response functions, large N limit.

**arXiv ID:** 2208.14686 **Authors:** Dustin Carrión-Ojeda, Hong Chen, Adrian El Baz, Sergio Escalera, Chaoyu Guan, Isabelle Guyon, Ihsan Ullah, Xin Wang, Wenwu Zhu **Published:** 2022-08-31T08:31:02Z **Abstract:** We present the design and baseline results for a new challenge in the ChaLearn meta-learning series, accepted at NeurIPS'22, focusing on "cross-domain" meta-learning. Meta-learning aims to leverage experience gained from previous tasks to solve new tasks efficiently (i.e., with bet...

**arXiv ID:** 2205.10937 **Authors:** Andrea Gesmundo, Jeff Dean **Published:** 2022-05-22T21:54:33Z **Abstract:** Most uses of machine learning today involve training a model from scratch for a particular task, or sometimes starting with a model pretrained on a related task and then fine-tuning on a downstream task. Both approaches offer limited knowledge transfer between different tasks, time-consuming human-driven customization to individual tasks and high computational costs especially wh...

Multi-Source Multi-View Graph Domain Adaptation with Hyperbolic Residual Encoding framework for cross-site Major Depressive Disorder (MDD) identification from resting-state fMRI. Use when working with multi-site neuroimaging data, cross-domain brain network analysis, heterogeneous functional connectivity views, or hyperbolic representation learning for clinical applications.

Skill for AI agent capabilities

Computational complexity lens for understanding how ML manages complex systems. Based on arxiv:2604.07233 'How Does Machine Learning Manage Complexity?' by Lance Fortnow. Use when analyzing ML's ability to model complex systems, understanding complexity bounds, P/poly-computable distributions, or when asked 'how does ML handle complexity?', 'ML complexity theory', 'computable distributions in ML'.

Energy-first neural architecture design framework based on biological principles. Systematic validation across vision, text, neuromorphic, and physiological datasets with 2,203 experiments. Activation: energy-first, neural architecture, biological principles, energy-regularized, lambda sweep.

... Activation: meta-learning, in-context learning, cross-subject

元学习上下文方法实现无需训练的跨被试脑解码。通过上下文学习实现训练无关的跨个体fMRI解码。适用于零样本脑解码、快速脑机接口、个体化神经科学。触发词：元学习脑解码、上下文学习、跨被试、训练无关、零样本。

Meta-Learning Biologically Plausible Plasticity Rules

Majorization lattice supermodularity and subadditivity framework — two structural majorization relations (precursors) underlying supermodularity and subadditivity of all sum-concave functions including Tsallis, Rényi, and Shannon entropies on the majorization lattice. Applies to quantum information theory, entropy inequalities, lattice theory. Activation: majorization, supermodularity, subadditivity, majorization lattice, Tsallis entropy, Rényi entropy, sum-concave, information theory lattice...

Statistical learning theory methodology proving majority-of-three voting is optimal in the realizable PAC setting

**arXiv ID:** 2303.16067 **Authors:** Aaron Pache, Mark CW van Rossum **Published:** 2023-03-26T16:17:04Z **Abstract:** When training neural networks for classification tasks with backpropagation, parameters are updated on every trial, even if the sample is classified correctly. In contrast, humans concentrate their learning effort on errors. Inspired by human learning, we introduce lazy learning, which only learns on incorrect samples. Lazy learning can be implemented in a few lines of code ...

**arXiv ID:** 2205.08836 **Authors:** Christoph Linse, Thomas Martinetz **Published:** 2022-05-18T10:08:28Z **Abstract:** Recent findings have shown that highly over-parameterized Neural Networks generalize without pretraining or explicit regularization. It is achieved with zero training error, i.e., complete over-fitting by memorizing the training data. This is surprising, since it is completely against traditional machine learning wisdom. In our empirical study we fortify these findings in ...

LARGE: Locally Adaptive Regularization for estimating Gaussian Graphical Models — improving brain network connectivity estimation via node-specific penalty tuning. Activation: Gaussian graphical model, GGM, graphical Lasso, GLASSO, brain connectivity, functional connectivity, precision matrix, adaptive regularization, network neuroscience.

Conserved Kinematic Representations for Zero-Shot Decoding in Handwriting BCIs. Methodology aligning neural activity to imagined kinematics for zero-shot capable ML decoding of unseen characters in BCI systems. Use when: researching brain-computer interfaces, motor cortex representations, zero-shot decoding, handwriting BCIs, kinematic primitives, logographic language neuroprosthetics, compositional motor control. Keywords: zero-shot BCI, kinematic representation, handwriting decoding, motor ...

Geometric analysis of attractor boundaries and storage capacity limits in kernel Hopfield networks trained with Kernel Logistic Regression (KLR). Covers attractor basin geometry, Ridge of Optimization, morphing analysis, SNR vs Cover's theorem, and dynamical stability. Activation: kernel Hopfield, KLR associative memory, attractor basin geometry, Ridge of Optimization, morphing analysis Hopfield, storage capacity Hopfield, SNR analysis Hopfield, Cover's theorem associative memory, crosstalk n...

Kernel Hopfield networks: geometric analysis of attractor boundaries and storage capacity limits. KLR-trained associative memories with P/N ~16 for random sequences and ~20 for structured data. Trigger words: kernel Hopfield, associative memory, KLR, attractor basin, storage capacity, kernel logistic regression.

Jeffreys Flow framework for robust Boltzmann generators and rare event sampling. Addresses mode collapse in multi-modal distributions using Jeffreys divergence + Parallel Tempering distillation. Use when: sampling rough energy landscapes, Boltzmann generators, rare events, quantum thermal states, path integral Monte Carlo, avoiding KL divergence mode collapse.

**arXiv ID:** 2102.09972 **Authors:** Noam Razin, Asaf Maman, Nadav Cohen **Published:** 2021-02-19T15:10:26Z **Abstract:** Recent efforts to unravel the mystery of implicit regularization in deep learning have led to a theoretical focus on matrix factorization -- matrix completion via linear neural network. As a step further towards practical deep learning, we provide the first theoretical analysis of implicit regularization in tensor factorization -- tensor completion via certain type of no...

**arXiv ID:** 2510.03744 **Authors:** Qianfei Fan, Jiayu Wei, Peijun Zhu, Wensheng Ye, Meie Fang **Published:** 2025-10-04T09:09:06Z **Abstract:** Accurate decade-scale daily runoff forecasting in small watersheds is difficult because signals blend drifting trends, multi-scale seasonal cycles, regime shifts, and sparse extremes. Prior deep models (DLinear, TimesNet, PatchTST, TiDE, Nonstationary Transformer, LSTNet, LSTM) usually target single facets and under-utilize unlabeled spans, limitin...

**arXiv ID:** 1902.01838 **Authors:** Amritanshu Agrawal, Wei Fu, Di Chen, Xipeng Shen, Tim Menzies **Published:** 2019-02-05T18:16:56Z **Abstract:** Machine learning techniques applied to software engineering tasks can be improved by hyperparameter optimization, i.e., automatic tools that find good settings for a learner's control parameters. We show that such hyperparameter optimization can be unnecessarily slow, particularly when the optimizers waste time exploring "redundant tunings"', i....

**arXiv ID:** 2512.04475 **Authors:** Timo Stoll, Chendi Qian, Ben Finkelshtein, Ali Parviz, Darius Weber, Fabrizio Frasca, Hadar Shavit, Antoine Siraudin, Arman Mielke, Marie Anastacio, Erik Müller, Maya Bechler-Speicher, Michael Bronstein, Mikhail Galkin, Holger Hoos, Mathias Niepert, Bryan Perozzi, Jan Tönshoff, Christopher Morris **Published:** 2025-12-04T05:30:31Z **Abstract:** Machine learning on graphs has made substantial progress across domains such as molecular property prediction a...