Data & Analytics

Bayesian decision-making framework for membership inference attacks on statistical releases using Bayesian network population models. Reframes membership inference with respect to populations represented as Bayesian networks, enabling more effective specialized attacks by incorporating prior information about attribute dependency structures. Use when analyzing statistical disclosure risk, designing membership inference attacks, or evaluating privacy of released statistics.

**arXiv ID:** 2301.02464 **Authors:** Vincenzo Lomonaco, Lorenzo Pellegrini, Gabriele Graffieti, Davide Maltoni **Published:** 2023-01-06T11:22:59Z **Abstract:** In recent years we have witnessed a renewed interest in machine learning methodologies, especially for deep representation learning, that could overcome basic i.i.d. assumptions and tackle non-stationary environments subject to various distributional shifts or sample selection biases. Within this context, several computational approa...

**arXiv ID:** 2407.05379 **Authors:** Maria Arostegi, Miren Nekane Bilbao, Jesus L. Lobo, Javier Del Ser **Published:** 2024-07-07T14:04:57Z **Abstract:** The ever-growing speed at which data are generated nowadays, together with the substantial cost of labeling processes cause Machine Learning models to face scenarios in which data are partially labeled. The extreme case where such a supervision is indefinitely unavailable is referred to as extreme verification latency. On the other hand, in...

**arXiv ID:** 2307.15092 **Authors:** Heng Zhang, Danilo Vasconcellos Vargas **Published:** 2023-07-27T05:20:20Z **Abstract:** Reservoir computing (RC), first applied to temporal signal processing, is a recurrent neural network in which neurons are randomly connected. Once initialized, the connection strengths remain unchanged. Such a simple structure turns RC into a non-linear dynamical system that maps low-dimensional inputs into a high-dimensional space. The model's rich dynamics, linear s...

**arXiv ID:** 1608.08905 **Authors:** Rajasekar Venkatesan, Meng Joo Er, Shiqian Wu, Mahardhika Pratama **Published:** 2016-08-31T15:14:06Z **Abstract:** In this paper, a novel extreme learning machine based online multi-label classifier for real-time data streams is proposed. Multi-label classification is one of the actively researched machine learning paradigm that has gained much attention in the recent years due to its rapidly increasing real world applications. In contrast to traditional...

**arXiv ID:** 2402.05144 **Authors:** Margaux Brégère, Julie Keisler **Published:** 2024-02-07T08:01:45Z **Abstract:** This work formulates model selection as an infinite-armed bandit problem, namely, a problem in which a decision maker iteratively selects one of an infinite number of fixed choices (i.e., arms) when the properties of each choice are only partially known at the time of allocation and may become better understood over time, via the attainment of rewards.Here, the arms are machi...

Secretary problem optimal stopping thresholds are exactly the convergents of 1/e via continued fractions. If p/q is a continued fraction convergent of 1/e with q at least 3, then for q applicants the optimal number to initially reject is p. Connects optimal stopping theory, continued fractions, and the mathematical constant e. Use when: optimal stopping problems, secretary problem analysis, continued fraction applications, 1/e thresholds, decision theory, sequential selection.

Supervised Deep Multimodal Matrix Factorization (SD3MF) methodology for interpretable brain network analysis. Generalizes SNMTF from unsupervised single-graph clustering to supervised prediction over populations of multimodal graphs. Learns deep hierarchical factorizations with shared latent representations that align subjects across modalities via encoder-decoder formulation. Use when: analyzing multimodal connectome data, building interpretable brain network classifiers, performing supervis...

SIMON: Saliency-aware Integrative Multi-view Object-centric Neural Decoding for zero-shot EEG-to-image retrieval. Uses foreground segmentation, saliency prediction, Saliency-Aware Sampling (SAS), and foveated multi-view integration to overcome center-bias limitations in EEG-to-image retrieval. Trigger words: saliency-aware EEG decoding, SIMON, EEG-to-image retrieval, foveated view, multi-view neural decoding, Saliency-Aware Sampling, object-centric neural decoding, zero-shot EEG image, THINGS...

SAE 最优性结构理论 - 解释 Sparse Autoencoders 如何从最优性条件提取可解释特征。涵盖层次分裂与吸收、残差结构、密集对立特征等现象的理论基础。

- **Title**: RTS Smoother-Guided Learning of Physics-Based Neural Differential Models - **Authors**: Ahmet Demirkaya, Georgios Stratis, Tales Imbiriba, Zachary D. Danziger, Deniz Erdogmus - **arXiv ID**: 2607.15180 - **URL**: http://arxiv.org/abs/2607.15180 - **Subjects**: Machine Learning (cs.LG); Systems and Control (eess.SY) - **Abstract**: Ordinary differential equations (ODEs) are widely used to model dynamical systems in physics, biology, neuroscience, and physiology, but in many applicati

RNN权重初始化、解的多样性与性能退化分析框架。研究不同初始化如何收敛到不同动力学解，分析网络规模、时间间隔、连接损伤对性能的优雅退化影响。适用于计算神经科学、RNN模型分析、脑皮层建模。触发词：RNN初始化、解多样性、性能退化、网络鲁棒性、优雅退化、weight initialization、degradation analysis、RNN dynamics、graceful degradation。

Paper analysis: Identifying structural design principles shaping computational abilities of recurrent neural networks. Demonstrates that local 2- and 3-cycles in connectivity strongly enhance computational ability of RNNs, and that adding sparse biologically-inspired interneurons dramatically increases capacity. Complete catalogs of network-function performance reveal most networks fail at most functions. Source: arXiv:2606.23874 (q-bio.NC, cs.NE), 2026-06-22. Activation keywords: RNN structu...

Bures-Wasserstein metric-based Riemannian self-attention network for robust EEG decoding

Associative presynaptic short-term plasticity via information-theoretic learning rules maximizing stimulus information under resource constraints

Renormalization group (RG) framework for analyzing scaling laws and criticality in brain activity. Connects 1/f noise, neuronal avalanches, and coarse-grained descriptions through RG theory. Activates: renormalization brain, scaling law neural activity, 1/f noise brain, neuronal avalanche scaling, coarse-graining neural dynamics, RG criticality brain, power law neural scaling.

Survival Reinforcement Learning (SRL) - online classification-based self-supervised RL that maximizes agent dwell time at target goals, extending survival value learning framework. Bypasses contrastive RL constraints and mitigates "bang-bang" control issues.

本地基座模型强化学习对齐工程实践 - 涵盖 RLHF/DPO/GRPO 算法选型、显存优化、框架选择、数据工程与全流程实施指南

Exact variance and Fano factor analytical formulae for arbitrary level crossings in stationary Gaussian processes. Extends the Kac-Rice mean crossing rate to capture clustering vs. regularity statistics, critical for neuronal spike train analysis, neural coding reliability, and stochastic neural dynamics. Use when analyzing spike train variability, threshold crossing statistics, or neural coding Fano factors.

CurveRL methodology — principled distribution-aware context reweighting for RLVR, using quantile coordinate transform where prompt weights depend on pass-rate rank and density rather than absolute values.

AGPO (Adaptive Group Policy Optimization) methodology — a critic-free refinement of GRPO that uses group-level statistics to adaptively control update magnitude and exploration. Uses a shared probe-derived statistical state to drive adaptive clipping (based on reward dispersion, skewness, probe entropy, policy entropy, KL drift) and bidirectional adaptive temperature sampling. Outperforms PPO/GRPO on 9 math/STEM benchmarks with Qwen2.5-14B. Use when: improving GRPO training stability, reducin...

Post-Recurrent Module (PRM) for explainable RNN-based P300 classification in BCIs — combines performance improvement with global/local explainability techniques for transparent EEG-based neural decoding. Activation triggers: PRM, P300 BCI, explainable RNN, EEG explainability, post-recurrent module, P300 classification, transparent BCI.

PRISM (Probabilistic Recurrent Intention Switching Model) methodology for multi-intention inverse reinforcement learning. Uses lightweight recurrent networks for intention switching with closed-form EM solution. Activation: 多意图 IRL, intention switching, PRISM, 目标切换, recurrent intention, EM algorithm.

PRISM (Probabilistic Recurrent Intention Switching Model) methodology for multi-intention inverse reinforcement learning. Uses lightweight recurrent networks for intention switching with closed-form EM solution. Activation: 多意图 IRL, intention switching, PRISM, 目标切换, recurrent intention, EM algorithm.