AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Paris: A Decentralized Trained Open-Weight Diffusion Model

Zhiying Jiang, Raihan Seraj, Marcos Villagra, Bidhan Roy

We present Paris, the first publicly released diffusion model pre-trained entirely through decentralized computation. Paris demonstrates that high-quality text-to-image generation can be achieved without centrally coordinated infrastructure. Paris is open for research and commercial use. Paris required implementing our Distributed Diffusion Training framework from scratch. The model consists of 8 expert diffusion models (129M-605M parameters each) trained in complete isolation with no gradient, parameter, or intermediate activation synchronization. Rather than requiring synchronized gradient updates across thousands of GPUs, we partition data into semantically coherent clusters where each expert independently optimizes its subset while collectively approximating the full distribution. A lightweight transformer router dynamically selects appropriate experts at inference, achieving generation quality comparable to centrally coordinated baselines. Eliminating synchronization enables training on heterogeneous hardware without specialized interconnects. Empirical validation confirms that Paris's decentralized training maintains generation quality while removing the dedicated GPU cluster requirement for large-scale diffusion models. Paris achieves this using 14$\times$ less training data and 16$\times$ less compute than the prior decentralized baseline.

arXiv

A Hamiltonian driven Geometric Construction of Neural Networks via the Lognormal family, Application to Financial Fraud Detection and to Network Security

Prosper Rosaire Mama Assandje, Landry Foka Marius, Arnaud Gires Fobasso Tchinda, Frédéric Barbaresco, Stéphane R. Gael Ekodeck, Serge Alain Ebele

We presents a method for constructing neural networks intrinsically on statistical manifolds via the lognormal distribution. We demonstrate this approach by formulating a neural network architecture directly on statistical manifold. The construction is driven by the Hamiltonian system that is equivalent to the gradient flow on this manifold. We define the network's input values using the coordinate system of this Hamiltonian dynamics, naturally embedded in the Poincar$\acute{e}$ disk. The core of our contribution lies in the derivation of the network's components from geometric principles: the rotation component of the synaptic weight matrix is determined by the Lie group action of $SU(1,1)$ on the disk, while the activation function emerges from the symplectic structure of the system. We subsequently obtain the complete weight matrix, including its translation vector, and the resulting output values.

arXiv

Towards White-Box Deep Wireless Sensing

Xie Zhang, Yina Wang, Chenshu Wu

The empirical success of deep learning has spurred its application to the radio-frequency (RF) domain, leading to significant advances in Deep Wireless Sensing (DWS). However, most existing DWS models remain black boxes, with ad-hoc architectures and learned representations lacking explicit physical and mathematical grounding, which limits their reliability and generalizability in real-world deployments. We present RF-CRATE, an early step towards white-box DWS grounded in the complex sparse rate reduction principle. Using the CR-Calculus framework, we derive a fully complex-valued transformer with mathematically interpretable self-attention and residual modules. To address labeled data scarcity, we introduce subspace regularization to enhance representation diversity, yielding a 19.98% average improvement. We evaluate RF-CRATE across heterogeneous RF modalities and human sensing tasks, including activity, gait, and gesture recognition, pose estimation, and respiration monitoring. Experiments on five datasets show that RF-CRATE remains competitive with strong black-box models while providing mathematically interpretable architectures and representations. Moreover, the complex-valued design achieves a 3.39% gain in classification accuracy and a 10.34% reduction in regression error. Our results demonstrate that mathematically grounded models can achieve strong performance in wireless sensing, offering a promising step towards physically aligned white-box DWS systems.

arXiv

AICO: Feature Significance Tests for Supervised Learning

Kay Giesecke, Enguerrand Horel, Chartsiri Jirachotkulthorn

Machine learning is central to modern science, industry, and policy, yet its predictive power often comes at the cost of transparency: we rarely know which input features truly drive a model's predictions. Without such understanding, researchers cannot draw reliable conclusions, practitioners cannot ensure fairness or accountability, and policymakers cannot trust or govern model-based decisions. Existing tools for assessing feature influence are limited; most lack statistical guarantees, and many require costly retraining or surrogate modeling, making them impractical for large modern models. We introduce AICO, a broadly applicable framework that turns model interpretability into an efficient statistical exercise. AICO tests whether each feature genuinely improves predictive performance by masking its information and measuring the resulting change. The method provides exact, finite-sample feature p-values and confidence intervals for feature importance through a simple, non-asymptotic hypothesis testing procedure. It requires no retraining, surrogate modeling, or distributional assumptions, making it feasible for large-scale algorithms. In both controlled experiments and real applications, from credit scoring to mortgage-behavior prediction, AICO reliably identifies the variables that drive model behavior, providing a scalable and statistically principled path toward transparent and trustworthy machine learning.

arXiv

ASVSim (AirSim for Surface Vehicles): A High-Fidelity Simulation Framework for Autonomous Surface Vehicle Research

Bavo Lesy, Siemen Herremans, Robin Kerstens, Jan Steckel, Walter Daems, Siegfried Mercelis, Ali Anwar

The transport industry has recently shown significant interest in unmanned surface vehicles (USVs), specifically for port and inland waterway transport. These systems can improve operational efficiency and safety, which is especially relevant in the European Union, where initiatives such as the Green Deal are driving a shift towards increased use of inland waterways. At the same time, a shortage of qualified personnel is accelerating the adoption of autonomous solutions. However, there is a notable lack of open-source, high-fidelity simulation frameworks and datasets for developing and evaluating such solutions. To address these challenges, we introduce AirSim for Surface Vehicles (ASVSim), an open-source simulation framework specifically designed for autonomous shipping research in inland and port environments. The framework combines simulated vessel dynamics with marine sensor simulation capabilities, including radar and camera systems and supports the generation of synthetic datasets for training computer vision models and reinforcement learning (RL) agents. Built upon Cosys-AirSim, ASVSim provides a comprehensive platform for developing autonomous navigation algorithms and generating synthetic datasets. The simulator supports research of both traditional control methods and deep learning-based approaches. Through experiments in waterway segmentation and autonomous navigation, we demonstrate the capabilities of the simulator in these research areas. ASVSim is provided as an open-source project under the MIT license, making autonomous navigation research accessible to a larger part of the ocean engineering community. See https://github.com/BavoLesy/ASVSim.

arXiv

StaQ: a Finite Memory Approach to Discrete Action Policy Mirror Descent

Alex Davey, Alena SHILOVA, Brahim Driss, Riad Akrour

In Reinforcement Learning (RL), regularization with a Kullback-Leibler divergence that penalizes large deviations between successive policies has emerged as a popular tool both in theory and practice. This family of algorithms, often referred to as Policy Mirror Descent (PMD), has the property of averaging out policy evaluation errors which are bound to occur when using function approximators. However, exact PMD has remained a mostly theoretical framework, as its closed-form solution involves the sum of all past Q-functions which is generally intractable. A common practical approximation of PMD is to follow the natural policy gradient or use actor-critic approaches, but this potentially introduces errors in the policy update. In this paper, we propose and analyze PMD-like algorithms for discrete action spaces that only keep the last $M$ Q-functions in memory. We show theoretically that for a finite and large enough $M$, an RL algorithm can be derived that introduces no errors from the policy update, yet keeps the desirable PMD property of averaging out policy evaluation errors. Using an efficient GPU implementation, we then show empirically on medium-scale RL benchmarks such as MinAtar that increasing $M$ improves performance up to a certain threshold after which the performance becomes close to that of exact PMD, reinforcing the theoretical findings that using an infinite sum might be unnecessary and that keeping in memory the last M Q-functions is a practical and theoretically grounded implementation of PMD.

arXiv

Reproducing Human Individual Motor Signatures: A Data-Driven Approach for Repetitive Motion

Angelo Di Porzio, Marco Coraggio

The deployment of autonomous virtual avatars (in extended reality) and robots in human group activities---such as rehabilitation therapy, sports, and manufacturing---is expected to increase as these technologies become more pervasive. Designing cognitive architectures and control strategies to drive these agents requires realistic models of human motion. Furthermore, recent research has shown that each person exhibits a unique velocity signature, highlighting how individual motor behaviors are both rich in variability and internally consistent. However, existing models only provide simplified descriptions of human motor behavior, hindering the development of effective cognitive architectures. In this work, we first show that motion amplitude provides a useful characterization of individual motor signatures, complementary to existing ones. Then, we propose a fully data-driven approach to generate original one-dimensional motion that captures the unique features of specific individuals, based on long short-term memory neural networks. We validate the architecture using real human data from participants performing spontaneous oscillatory motion. Thorough statistical analyses support that our model reproduces the velocity distribution and amplitude envelopes of the individual it was trained on, while remaining distinct from others.

arXiv

Dimensionality reduction for homological stability and global structure preservation

Alexander Kolpakov, Igor Rivin

We propose DiRe, a force-directed dimensionality reduction framework designed to preserve global structure and homological features while remaining practical on modern hardware. The method combines an initial embedding with a graph-based layout optimization and evaluates the resulting low-dimensional representation using local distortion, context preservation, and persistent homology measures. Across the benchmark suite considered here, DiRe provides a complementary tradeoff to UMAP and tSNE: it is designed less as a purely local visualization heuristic and more as a framework for embeddings whose large-scale geometry can be quantified through Betti curves and persistence diagrams.

arXiv

Dual-Force: Enhanced Offline Diversity Maximization under Imitation Constraints

Pavel Kolev, Marin Vlastelica, Georg Martius

Offline diversity maximization under imitation constraints can transform demonstration data into a set of distinct behavioral policies, improving robustness to distribution shift without additional environment interaction. In practice, however, existing offline approaches often rely on mutual-information objectives that require training a skill discriminator and can become unstable under the non-stationary rewards induced by alternating Lagrangian optimization. We introduce Dual-Force, an offline algorithm that (i) maximizes diversity using an off-policy estimator of a Van der Waals (VdW) force objective computed from successor features, eliminating the skill discriminator, and (ii) stabilizes training under non-stationary intrinsic rewards by conditioning the value function and policy on a pre-trained Functional Reward Encoding (FRE). The FRE code also enables zero-shot recall of every encountered skill via its associated latent representation, removing the need to pre-specify a fixed number of skills. On two Solo12 simulation benchmarks (locomotion and obstacle navigation), Dual-Force recovers diverse high-performing behaviors while matching a target expert state occupancy and improves robustness in adversarial obstacle variations.

arXiv

Unified continuous-time q-learning for mean-field game and mean-field control problems

Xiaoli Wei, Xiang Yu, Fengyi Yuan

This paper studies the continuous-time q-learning in mean-field jump-diffusion models in a setting where the environment simulator does not provide direct access to the population distribution. We propose the integrated q-function in decoupled form (decoupled Iq-function) and establish its martingale characterization, which provides a unified policy evaluation rule for both mean-field game (MFG) and mean-field control (MFC) problems. Moreover, we consider the learning procedure where population distribution is updated based on the representative agent's state values. Depending on the task to solve the MFG or MFC problem, we can employ the decoupled Iq-function differently to characterize the mean-field equilibrium policy or the mean-field optimal policy respectively. Based on these theoretical findings, we devise a unified parametric q-learning algorithm for both MFG and MFC problems by utilizing test policies and the averaged martingale orthogonality condition. In two applications within and beyond LQ framework, we illustrate the effectiveness and efficiency of our unified parametric q-learning algorithm for both MFG and MFC learning tasks.

arXiv

On the Expressive Power of Sparse Geometric MPNNs

Yonatan Sverdlov, Nadav Dym

Motivated by applications in chemistry and other sciences, we study the expressive power of message-passing neural networks for geometric graphs, whose node features correspond to 3-dimensional positions. Recent work has shown that such models can separate generic pairs of non-isomorphic geometric graphs, though they may fail to separate some rare and complicated instances. However, these results assume a fully connected graph, where each node possesses complete knowledge of all other nodes. In contrast, often, in application, every node only possesses knowledge of a small number of nearest neighbors. This paper shows that generic pairs of non-isomorphic geometric graphs can be separated by message-passing networks with rotation equivariant features as long as the underlying graph is connected. When only invariant intermediate features are allowed, generic separation is guaranteed for generically globally rigid graphs. We introduce a simple architecture, EGENNET, which achieves our theoretical guarantees and compares favorably with alternative architecture on synthetic and chemical benchmarks. Our code is available at https://github.com/yonatansverdlov/E-GenNet.

arXiv

Communication-Efficient Secure Aggregation in Decentralized Learning

Sayan Biswas, Anne-Marie Kermarrec, Rafael Pires, Rishi Sharma, Milos Vujasinovic

Decentralized learning (DL) enables participants to collaboratively train models without a central server, yet it faces significant scalability challenges that demand sparsification to reduce the prohibitive communication costs of peer-to-peer exchange. While secure aggregation effectively mitigates privacy risks in standard settings, it has remained fundamentally incompatible with sparsification in decentralized networks due to the mismatch of indices across local updates, forcing a trade-off between communication efficiency and privacy. This paper introduces CESAR, a novel protocol that resolves this incompatibility by integrating secure aggregation and sparsification to provide provable defense against honest-but-curious and colluding adversaries. By coordinating masks over parameter intersections, CESAR supports node dropouts and robust privacy without central aggregation. Empirical evaluations on models up to 124 million parameters demonstrate that CESAR matches the accuracy of non-private baselines while cutting total data exchange by 66.7 % compared to a standard full-parameter decentralized protocol (D-PSGD). With TopK sparsification on IID data, CESAR even exceeds by 0.3 % the accuracy achieved by D-PSGD with sparsification. Collectively, these results establish CESAR as the first decentralized protocol to achieve both privacy and communication efficiency through secure aggregation in DL.

arXiv

Nonparametric Partial Disentanglement via Mechanism Sparsity: Sparse Actions, Interventions and Sparse Temporal Dependencies

Sébastien Lachapelle, Pau Rodríguez López, Yash Sharma, Katie Everett, Rémi Le Priol, Alexandre Lacoste, Simon Lacoste-Julien

This work introduces a novel principle for disentanglement we call mechanism sparsity regularization, which applies when the latent factors of interest depend sparsely on observed auxiliary variables and/or past latent factors. We propose a representation learning method that induces disentanglement by simultaneously learning the latent factors and the sparse causal graphical model that explains them. We develop a nonparametric identifiability theory that formalizes this principle and shows that the latent factors can be recovered by regularizing the learned causal graph to be sparse, under some assumptions such as the absence of instantaneous causal effects between latent factors. More precisely, we show identifiability up to a novel equivalence relation we call consistency, which allows some latent factors to remain entangled (hence the term partial disentanglement). To describe the structure of this entanglement, we introduce the notions of entanglement graphs and graph preserving functions. We further provide a graphical criterion which guarantees complete disentanglement, that is identifiability up to permutations and element-wise transformations. We demonstrate the scope of the mechanism sparsity principle as well as the assumptions it relies on with several worked out examples. For instance, the framework shows how one can leverage multi-node interventions with unknown targets on the latent factors to disentangle them. We further draw connections between our nonparametric results and the now popular exponential family assumption. Lastly, we propose an estimation procedure based on variational autoencoders and a sparsity constraint and demonstrate it on various synthetic datasets. This work is meant to be a significantly extended version of a work published at CLeaR 2022.

arXiv

Information Processing by Neuron Populations in the Central Nervous System: A Theory of the Mathematical Structure of Data and Operations

Martin N. P. Nilsson

In the mammalian central nervous system, neurons are organized into populations communicating by spike trains propagating along axonal bundles. How such populations encode and transform information is only partially understood. In this study we introduce a mathematical framework derived from a mechanistic model of a single plastic neuron. Within this framework, an algebra of convex cones can rigorously characterize population-level activity. This algebra provides a natural language describing information representation and processing. Neuron populations are thereby interpreted not as passive transmitters but as operators acting within this algebraic structure. When interconnected, such populations realize compact algebraic expressions whose functional repertoire includes specialization, generalization, novelty detection, dimensionality reduction, inverse modeling, prediction, and associative memory. Finally, the approach highlights the role of matrix embeddings in extending representational capacity beyond that afforded by vector-based models. In particular, such embeddings support hierarchical concept formation and structured information processing, with potential implications for both cognitive neuroscience and artificial intelligence. This paper assumes familiarity with elementary functional analysis and algebras of operators.

arXiv

Beyond Black-Box Advice: Learning-Augmented Algorithms for MDPs with Q-Value Predictions

Tongxin Li, Yiheng Lin, Shaolei Ren, Adam Wierman

We study the tradeoff between consistency and robustness in the context of a single-trajectory time-varying Markov Decision Process (MDP) with untrusted machine-learned advice. Our work departs from the typical approach of treating advice as coming from black-box sources by instead considering a setting where additional information about how the advice is generated is available. We prove a first-of-its-kind consistency and robustness tradeoff given Q-value advice under a general MDP model that includes both continuous and discrete state/action spaces. Our results highlight that utilizing Q-value advice enables dynamic pursuit of the better of machine-learned advice and a robust baseline, thus result in near-optimal performance guarantees, which provably improves what can be obtained solely with black-box advice.

arXiv

POSSE-kNN: Pathwise Out-of-Bag Selected Subspace Ensembles for Binary Classification

Zardad Khan, Amjad Ali, Najd Adeed, Saeed Aldahmani

Nearest neighbour classification is attractive for tabular data, but its performance can deteriorate when a fixed query centred neighbourhood does not follow the local class geometry. This study evaluates POSSE-$k$NN, a pathwise $k$ nearest neighbour ensemble that combines bootstrap sampling, random feature subspaces, out-of-bag (OOB) screening, and selective voting. Within each randomized candidate, pathwise selection first chooses the training observation nearest to the query and then chooses each subsequent neighbour relative to the observation accepted at the preceding step. After the candidate is fully specified, its OOB error is computed. Five hundred candidates are generated, ranked by OOB error, and the best 25% are retained. The method is evaluated on ten binary benchmark datasets using repeated 70/30 train/test partitions and six established comparators. Across the dataset level means, POSSE-kNN attains an accuracy of 0.740, Cohen's kappa of 0.412, and a Brier score of 0.175, giving the best aggregate result for all three criteria. It has the highest unrounded mean accuracy and kappa on eight datasets; RkNN and SVM lead on the other two. A neighbourhood size analysis on three datasets shows stable behaviour for k={3, 5, 7} when the path geometry is informative and identifies a case in which alternative neighbourhood rules are preferable.

arXiv

Tensor Data Scattering and the Impossibility of Slicing Theorem

Wuming Pan

This paper proposes a standard way to represent sparse tensors. A broad theoretical framework for tensor data scattering methods used in various deep learning frameworks is established. This paper presents a theorem that is very important for performance analysis and accelerator optimization for implementing data scattering. The theorem shows how the impossibility of slicing happens in tensor data scattering. A sparsity measuring formula is provided, which can effectively indicate the storage efficiency of sparse tensor and the possibility of parallelly using it. A Python reference implementation is provided as ancillary material with this arXiv submission.

arXiv

TEXAS: Task-Expert-Aware Supervision for Downstream Mixture-of-Experts LLM Adaptation

Guanzhi Deng, Haibo Wang, Kuan Wu, Xiangru Jian, Shing Yin Wong, Sichun Luo, Zhuoran Wang, Linqi Song

Mixture-of-Experts (MoE) language models route each token through a small subset of experts, making routing patterns useful for identifying task-relevant experts during downstream adaptation. Yet current approaches have two limitations: task experts are typically identified from aggregate routing statistics that reflect usage rather than association with successful task completion, and task-expert activations remain underexplored as signals for supervision allocation. We introduce Task-Expert-Aware Supervision (TEXAS), which combines correctness-conditioned task expert discovery with token-level supervision allocation. TEXAS compares expert activations on instances that the base model solves successfully and those it fails to solve, and retains experts more strongly activated on successful instances. During fine-tuning, it upweights answer tokens in failed instances when they activate these experts. TEXAS therefore leverages existing routing behavior without restricting adaptation to a fixed expert subset or imposing an explicit target routing distribution. Across three MoE models and six benchmarks, TEXAS achieves the best or tied-best performance in 17 of 18 settings and improves over the strongest baseline by 1.3--1.5 points on average. Ablations and further analyses validate both the discovered experts and the resulting supervision strategy.

arXiv

Position: It's Time to Optimize LLMs for Self-Consistency

Itamar Pres, Belinda Z. Li, Laura Ruis, Zifan Carl Guo, Keya Hu, Mehul Damani, Isha Puri, Ekdeep Singh Lubana, Jacob Andreas

Despite ever-increasing sophistication in language model (LM) pre- and post-training pipelines, many important failures persist: models overcondition on user framing ("sycophancy"), exhibit incomplete logical generalization, and produce confident but incorrect responses. We argue that these failures arise from a modeling assumption permeating all aspects of the pipeline: that behavior can be specified and evaluated independently on single-output pairs. Many model failures are difficult, if not impossible, to detect without reasoning about relationships between a model's responses across inputs. In this position paper, we propose self-consistency as a framework for understanding these failures. We first observe that a wide variety of techniques designed to improve specific aspects of LM behavior-targeting properties as diverse as adversarial robustness and factual coherence-can be understood as special cases of a common "consistency optimization" procedure and addressed with a standard set of optimization tools. We next outline a set of new model properties that could be achieved by optimizing for consistency, and conclude with a discussion of what it would mean to develop generally consistent LMs, including the capabilities they would enable and the objections they raise.

arXiv

ORCA: ORgan-Centroid Aggregation for Training-Free 3D CT Visual Token Compression

Renjie Liang, Zijian Xu, Jinqian Pan, Chengkun Sun, Zhengkang Fan, Shawn Li, You Qin, Mei Liu, Jie Xu

A 3D CT scan entering a vision-language model produces a long sequence of visual tokens, often thousands to tens of thousands per volume, and this sequence must be compressed before a language model can consume it. Token compression is well studied in general vision, but little of it targets 3D CT specifically. A common baseline is grid average, which pools regular grid cells and can blend distinct anatomy, lesion, and air into one token. We present \textbf{ORCA} (ORgan-Centroid Aggregation), a token compressor for 3D CT. It merges adjacent tokens with organ guidance and adds a sinusoidal encoding of each region's centroid to preserve spatial layout. This preserves the anatomical information a downstream model needs. ORCA is training-free and plug-and-play, producing an adjustable token set without any model change or text query. We evaluate it across two datasets (CT-RATE and Merlin) and five encoders. The evaluation spans two task types: attribute prediction over five families (size, density, location, texture, and disease) and text generation (visual question answering and report generation). At matched token budgets, ORCA improves consistently over existing compression methods. It shrinks the visual context $64\times$ and its KV-cache $50\times$, and is $31\times$ faster to process each volume. Code released at https://github.com/renjie-liang/ORCA-3DCT.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.