Skip to yearly menu bar Skip to main content


Poster

DINOv3

Oriane Siméoni ⋅ Huy V. Vo ⋅ Maximilian Seitzer ⋅ Federico Baldassarre ⋅ Maxime Oquab ⋅ Cijo Jose ⋅ Vasil Khalidov ⋅ Marc Szafraniec ⋅ Seung Eun Yi ⋅ Michaël Ramamonjisoa ⋅ Francisco Massa ⋅ Daniel Haziza ⋅ Luca Wehrstedt ⋅ Jianyuan Wang ⋅ Timothée Darcet ⋅ Théo Moutakanni ⋅ Leonel Sentana ⋅ Claire Roberts ⋅ Andrea Vedaldi ⋅ Jamie Tolan ⋅ John Brandt ⋅ Camille Couprie ⋅ Julien Mairal ⋅ Herve Jegou ⋅ Patrick Labatut ⋅ Piotr Bojanowski
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P277
Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This technical report introduces DINOv3, a major milestone toward realizing this vision by leveraging simple yet effective strategies. First, we leverage the benefit of scaling both dataset and model size by careful data preparation, design, and optimization. Second, we introduce a new method called Gram anchoring, which effectively addresses the known yet unsolved issue of dense feature maps degrading during long training schedules. Finally, we apply post-hoc strategies that further enhance our models' flexibility with respect to resolution, model size, and alignment with text. As a result, we present a versatile vision foundation model that outperforms the specialized state of the art across a broad range of settings, without fine-tuning. DINOv3 produces high-quality dense features that achieve outstanding performance on various vision tasks, significantly surpassing previous self- and weakly-supervised foundation models. We also share the DINOv3 suite of vision models, designed to advance the state of the art on a wide spectrum of tasks and data by providing scalable solutions for diverse resource constraints and deployment scenarios.
Show more
View full details
Poster

MatchEx: Model-Level GNN Explanations with Multi-Granular Insights

Sayan Saha ⋅ Sanghamitra Bandyopadhyay
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P310
Graph Neural Networks (GNNs) are increasingly deployed in high-stakes domains where interpretability is crucial. Existing model-level explanation methods largely rely on generative models, which often produce motifs that fail to resemble real instances, cannot account for the diversity of discriminative motifs recognized by the classifier for a target class and lack mechanisms for translating global explanations to instance-level insights. We present MatchEx, a framework that discovers discriminative motifs directly from real instances by optimizing a novel matching objective. Unlike isomorphism, which can only recover identical motifs that rarely occur in real-world graphs, this objective extends beyond exact matches to provably recover semantically similar motifs, allowing generalizable explanations. The matching mechanism also enables projection of class level rationales onto individual graphs for faithful instance-level insights. When a single motif fails to explain all instances, MatchEx adaptively partitions the instances in a class into coherent subgroups with distinct rationales. Extensive experiments across six real and synthetic datasets show that MatchEx consistently outperforms state-of-the-art baselines, delivering coherent, generalizable, and multi-granular explanations.
Show more
View full details
Poster

Gradient Span Algorithms Make Predictable Progress in High Dimension

Felix Benning ⋅ Leif Döring
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P158
We prove that all 'gradient span algorithms' have asymptotically deterministic behavior on scaled Gaussian random functions as the dimension tends to infinity. This is a functional generalization of similar results for random quadratic functions and spin glasses. They explain the counterintuitive phenomenon that different training runs of many large machine learning models result in approximately equal cost curves despite random initialization on a complicated non-convex landscape. This 'predictable progress' phenomenon is exploited by the AutoML community: Since the optimization progress of a single run is already representative, multiple retries with the same hyperparameters are not necessary.
Show more
View full details
Poster

AdaCubic: An Adaptive Cubic Regularization Optimizer for Deep Learning

Ioannis Tsingalis ⋅ Constantine Kotropoulos ⋅ Corentin Briat
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P155
A novel regularization technique, AdaCubic, is proposed that adapts the weight of the cubic term. The heart of AdaCubic is an auxiliary optimization problem with cubic constraints that dynamically adjusts the weight of the cubic term in Newton’s cubic regularized method. We use Hutchinson’s method to approximate the Hessian matrix, thereby reducing computational cost. We demonstrate that AdaCubic inherits the cubically regularized Newton method’s local convergence guarantees. Our experiments in Computer Vision, Natural Language Processing, and Signal Processing tasks demonstrate that AdaCubic outperforms or competes with several widely used optimizers. Unlike other adaptive algorithms that require hyperparameter fine-tuning, AdaCubic is evaluated with a fixed set of hyperparameters, rendering it a highly attractive optimizer in settings where fine-tuning is infeasible. This makes AdaCubic an attractive option for researchers and practitioners alike. To our knowledge, AdaCubic is the first optimizer to leverage cubic regularization in scalable deep learning applications.
Show more
View full details
Poster

Fast Wasserstein rates for estimating probability distributions of probabilistic graphical models

Daniel Bartl ⋅ Stephan Eckstein
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P110
Using i.i.d. data to estimate a high-dimensional distribution in Wasserstein distance is a fundamental instance of the curse of dimensionality. We explore how structural knowledge about the data-generating process which gives rise to the distribution can be used to overcome this curse. More precisely, we work with the set of distributions of probabilistic graphical models for a given directed acyclic graph. It turns out that this knowledge is only helpful if it can be quantified, which we formalize via smoothness conditions on the transition kernels in the disintegration corresponding to the graph. In this case, we prove that the rate of estimation is governed by the local structure of the graph, more precisely by dimensions corresponding to single nodes together with their parent nodes. The precise rate depends on the exact notion of smoothness assumed for the kernels, where either weak (Wasserstein-Lipschitz) or strong (bidirectional Total-Variation-Lipschitz) conditions lead to different results. We prove sharpness under the strong condition and show that this condition covers, as a special case, distributions having a positive Lipschitz density.
Show more
View full details
Poster

Many Circuits, One Mechanism: Input Variation and Evaluation Granularity in Circuit Discovery

Alireza Bayat Makou ⋅ Jingcheng Niu ⋅ Subhabrata Dutta ⋅ Iryna Gurevych
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P185
Circuit discovery methods identify subgraphs that explain specific model behaviors, and structural differences between discovered circuits are commonly interpreted as evidence of distinct mechanisms. We test this assumption by drawing input tokens from bands defined by their frequency in the pretraining data while holding the task fixed. The discovered circuits appear specialized by frequency when compared structurally, but functional and representational analyses show no reliable evidence of corresponding differences in their computations. We term this mismatch phantom specialization. Using the Literal Sequence Copying task across four frequency bands plus a control sampled according to token frequency, we extract 75 circuits from five Pythia models (70M-1.4B). We find that structurally distinct circuits implement the same computation: band-specific edges transfer broadly across bands, a core shared across most bands recovers at least 99% of circuit performance in models above 70M, and causal interchange interventions confirm that internal representations are interchangeable across frequency bands. A smaller replication on a subject-verb agreement task shows the same pattern: circuits differ structurally, transfer broadly across bands, and the core shared by most bands recovers nearly all of the circuit's accuracy in all five models. Repeated extractions within the same frequency band further suggest that discovery algorithms sample from an equivalence class of valid subgraphs rather than recovering a unique mechanism. Standard evaluation practice obscures this pattern: source-level evaluation inflates apparent faithfulness, while edge-level evaluation reveals the many-to-one mapping from structure to function. We find no reliable evidence that frequency bands are processed by different computations. In this work, we do not test specialization at the level of token positions or of features inside a component. Our results show that structural differences between circuits are not sufficient evidence for distinct mechanisms, and that exposing this requires edge-level evaluation and cross-condition transfer tests.
Show more
View full details
Poster

TropNNC: Structured Neural Network Compression Using Tropical Geometry

Konstantinos Fotopoulos ⋅ Petros Maragos ⋅ Panagiotis Misiakos
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P245
We present TropNNC, a framework for compressing neural networks with linear and convolutional layers and ReLU-type activations using tropical geometry. By representing a network’s output as a tropical rational function, TropNNC enables structured compression via reduction of the corresponding tropical polynomials. Our method identifies redundancy via similarity and improves upon the geometric approximation of previous work by adaptively selecting the weights of retained neurons. We relate it to SVD and spectral clustering, and provide insights into network compression beyond the specific setting considered. We provide the tightest known theoretical compression bound, and the first successful application of tropical geometry to convolutional layers. TropNNC requires access only to network weights – no training data – and achieves competitive performance on MNIST, CIFAR, and ImageNet, matching strong baselines such as ThiNet and CUP.
Show more
View full details
Poster

ARC-Encoder: learning compressed text representations for large language models

Hippolyte Pilchen ⋅ Edouard Grave ⋅ Patrick Perez
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P290
Recent techniques such as retrieval-augmented generation or chain-of-thought reasoning have led to longer contexts and increased inference costs. Context compression techniques can reduce these costs, but the most effective approaches require fine-tuning the target model or even modifying its architecture. This can degrade its general abilities when not used for this specific purpose. Here we explore an alternative approach: an encoder that compresses the context into continuous representations which replace token embeddings in decoder LLMs. First, we perform a study of training strategies and architecture choices for the encoder. Our findings led to the design of an Adaptable text Representations Compressor, named ARC-Encoder, which outputs $x$ times fewer continuous representations (typically $x\!\in\!\{4,8\}$) than text tokens. We evaluate ARC-Encoder across a variety of LLM usage scenarios, ranging from in-context learning to context window extension, on both instruct and base decoders. Results show that ARC-Encoder achieves strong performance on several benchmarks and tasks while improving computational efficiency at inference. Finally, we demonstrate that our models can be adapted to multiple decoders simultaneously, allowing a single encoder to generalize across different decoder LLMs. This makes ARC-Encoder a flexible and efficient solution for portable encoders that can support multiple LLMs with only small MLPs.
Show more
View full details
Poster

Privacy Guarantees in Posterior Sampling under Contamination

Shenggang Hu ⋅ Hongsheng Dai ⋅ Louis Aslett ⋅ Murray Pollock ⋅ Gareth Roberts
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P203
In recent years, differential privacy has been adopted by tech-companies and governmental agencies as the standard for measuring privacy in algorithms. In this article, we study differential privacy in Bayesian posterior sampling settings. We begin by considering differential privacy in the most common privatisation setting in which Laplace or Gaussian noise is injected into the output. In an effort to achieve better differential privacy, we consider adopting {\em Huber's contamination model} for use within privacy settings, and replace at random data points with samples from a heavy-tailed distribution ({\em instead} of injecting noise into the output). We derive bounds for the differential privacy level $(\epsilon,\delta)$ of our approach, without requiring bounded observation and parameter spaces, a restriction commonly imposed in the literature. We further consider for our approach the effect of sample size on the privacy level and the rate at which $(\epsilon,\delta)$ converges to zero. Asymptotically, our contamination approach is fully private with no information loss. We also provide examples of inference models for which our approach applies, with theoretical convergence rate analysis and simulation studies.
Show more
View full details
Poster

Exploring the Limitations of Layer Synchronization in Spiking Neural Networks

Roel Koopman ⋅ Amirreza Yousefzadeh ⋅ Mahyar Shahsavari ⋅ Guangzhi Tang ⋅ Manolis Sifalakis
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P256
Neural-network processing in machine learning applications relies on layer synchronization. This is practiced even in artificial Spiking Neural Networks (SNNs), which are touted as consistent with neurobiology, in spite of processing in the brain being in fact asynchronous. A truly asynchronous system however would allow all neurons to evaluate concurrently their threshold and emit spikes upon receiving any presynaptic current. Omitting layer synchronization is potentially beneficial, for latency and energy efficiency, but asynchronous execution of models previously trained with layer synchronization may entail a mismatch in network dynamics and performance. We present and quantify this problem, and show that models trained with layer synchronization either perform poorly in absence of the synchronization, or fail to benefit from any energy and latency reduction, when such a mechanism is in place. We then explore a potential solution direction, based on a generalization of backpropagation-based training that integrates knowledge about an asynchronous execution scheduling strategy, for learning models suitable for asynchronous processing. We experiment with 2 asynchronous neuron execution scheduling strategies in datasets that encode spatial and temporal information, and we show the potential of asynchronous processing to use less spikes (up to 50%), complete inference faster (up to 2x), and achieve competitive or even better accuracy (up to 10% higher). Our exploration affirms that asynchronous event-based AI processing can be indeed more efficient, but we need to rethink how we train our SNN models to benefit from it. (Source code available at: https://github.com/RoelMK/asynctorch)
Show more
View full details
Poster

Sublinear Variational Optimization of Gaussian Mixture Models with Millions to Billions of Parameters

Sebastian Salwig ⋅ Till Kahlke ⋅ Florian Hirschberger ⋅ Dennis Forster ⋅ Jörg Lücke
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P285
Gaussian Mixture Models (GMMs) range among the most frequently used models in machine learning. However, training large, general GMMs becomes computationally prohibitive for data sets that have many data points N of high-dimensionality D. For GMMs with arbitrary covariances, we here derive a highly efficient variational approximation, which is then integrated with mixtures of factor analyzers (MFAs). For GMMs with C components, our proposed algorithm substantially reduces runtime complexity from O(NCD²) per iteration to a complexity scaling linearly with D and sublinearly with NC. In numerical experiments, we first validate that the complexity reduction results in a sublinear scaling for the entire GMM optimization process. Second, we show on large-scale benchmarks that the sublinear algorithm results in speed-ups of an order-of-magnitude compared to the state-of-the-art. Third, as a proof of concept, we finally train GMMs with over 10 billion parameters on about 100 million images, observing training times of less than nine hours on a single state-of-the-art CPU. Finally, and fourth, we demonstrate the effectiveness of large-scale GMMs on the task of zero-shot image denoising, where sublinear training results in state-of-the-art denoising times while competitive denoising performance is maintained.
Show more
View full details
Poster

An Anytime Algorithm for Good Arm Identification

Marc Jourdan ⋅ Andrée Delahaye-Duriez ⋅ Clémence Réda
Dec 11, 1:00 AM - 3:00 AM Paris Poster Hall #P121
In good arm identification (GAI), the goal is to identify one arm whose average performance exceeds a given threshold, referred to as a good arm, if it exists. Few works have studied GAI in the fixed-budget setting when the sampling budget is fixed beforehand, or in the anytime setting, when a recommendation can be asked at any time. We propose APGAI, an anytime and parameter-free sampling rule for GAI in stochastic bandits. APGAI can be straightforwardly used in fixed-confidence and fixed-budget settings. First, we derive upper bounds on its probability of error at any time. They show that adaptive strategies can be more efficient in detecting the absence of good arms than uniform sampling in several diverse instances. Second, when APGAI is combined with a stopping rule, we prove upper bounds on the expected sampling complexity, holding at any confidence level. Finally, we show the good empirical performance of APGAI on synthetic and real-world data. Our work offers an extensive overview of the GAI problem in all settings.
Show more
View full details
Poster

Regret minimization in Linear Bandits with offline data via extended D-optimal exploration.

Sushant Vijayan ⋅ Arun Suggala ⋅ Karthikeyan Shanmugam ⋅ Soumyabrata Pal
Dec 11, 1:00 AM - 3:00 AM Paris Poster Hall #P159
We consider the problem of online regret minimization in stochastic linear bandits with access to prior observations (\emph{i.e.,} offline data) from the underlying bandit model. This setting is highly relevant to numerous applications where extensive offline data is often available, such as in recommendation systems, personalized healthcare, and online advertising. Consequently, this problem has been studied intensively in recent works such as~\cite{banerjee2022artificial, wagenmaker2022leveraging, agrawal2023optimal,hao2023leveraging,cheung2024leveraging}. We introduce the Offline-Online Phased Elimination (\oope) $\text{ }$algorithm, that effectively incorporates the offline data to substantially reduce the online regret compared to prior work. To leverage offline information prudently, \oope$\text{ }$ uses an extended D-optimal design within each exploration phase. We show that \oope $\text{ }$achieves an online regret is $\tilde{O}(\sqrt{d_{\text{eff}} T \log \left(|\mathcal{A}|T\right)}+d^2)$, where $\mathcal{A}$ is the action set, $d$ is the dimension and $T$ is the online horizon. $d_{\text{eff}}\hspace{0.1cm} (\leq d)$ is the \emph{effective problem dimension} which measures the number of poorly explored directions in offline data and depends on the eigen-spectrum $(\lambda_k)_{k \in [d]}$ of the Gram matrix of the offline data. Thus the eigen-spectrum $(\lambda_k)_{k \in [d]}$ is a quantitative measure of the \emph{quality} of offline data. If the offline data is poorly explored ($d_{\text{eff}}\approx d$), we recover the established regret bounds for purely online linear bandits. Conversely, when offline data is abundant ($T_{\text{off}} \gg T$) and well-explored ($d_{\text{eff}} = o(1) $), the online regret reduces substantially. Additionally, we provide the first known minimax regret lower bounds in this setting that depend explicitly on the quality of the offline data. These lower bounds establish the optimality of our algorithm \footnote{Optimal within log factors in $T, T_{\text{off}}$ and additive constants in $d$} in regimes where offline data is either well-explored or poorly explored. Finally, by using a Frank-Wolfe approximation to the extended optimal design we further improve the $O(d^{2})$ term to $O\left(\frac{d^{2}}{d_{\text{eff}}} \min \{ d_{\text{eff}},1\} \right)$, which can be substantial in high dimensions with moderate quality of offline data $d_{\text{eff}} = \Omega(1)$.
Show more
View full details
Poster

Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning

Charlotte Beylier ⋅ Hannah Selder ⋅ Arthur Fleig ⋅ Simon M. Hofmann ⋅ Nico Scherf
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P197
The emergence and evolution of feature reliance in deep reinforcement learning agents remain poorly understood. Here, we introduce a methodological framework for analyzing the learning process through quantitative analysis of saliency maps. This approach aggregates saliency information at the object and modality level into hierarchical attention profiles, quantifying how agents allocate attention over time, thereby forming attention trajectories throughout training. These profiles are then compared across controlled conditions, connected to behavioral measurements and reproduced with different saliency methods to assess the robustness of the findings. Applied to Atari 2600 benchmarks, custom Pong environments, and biomechanical user simulations in visuomotor tasks, this framework uncovers algorithm-specific attention biases, diagnosed unintended reward-driven strategies, and overfitting to redundant sensory channels. These patterns correspond to measurable behavioral differences, demonstrating empirical links between attention profiles, learning dynamics, and agent behavior. The results establish attention trajectories as a promising diagnostic axis for tracing how feature reliance develops during training and for identifying biases and vulnerabilities invisible to performance metrics alone.
Show more
View full details
Poster

Think2SQL: Blueprinting Reward Density and Advantage Scaling for Effective Text-to-SQL Reasoning

Simone Papicchio ⋅ Simone Rossi ⋅ Luca Cagliero ⋅ Papotti Paolo
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P143
While Large Language Models (LLMs) have advanced the state-of-the-art in Text-to-SQL, robust reasoning in complex, multi-table environments remains a bottleneck for parameter-efficient models. This paper presents a systematic empirical study on injecting reasoning capabilities into Text-to-SQL through the lens of Reinforcement Learning with Verifiable Rewards (RLVR) for the Qwen3 model family. We uncover a critical interplay between reward density, advantage scaling, and model capacity. Our analysis yields four primary insights. First, we propose a novel execution-guided dense reward function that significantly outperforms binary signals and existing state-of-the-art rewards by providing granular feedback at the instance level. Second, we analyze the mechanics of advantage calculation, demonstrating that while large models thrive on sparse signals with aggressive advantage scaling, smaller models require dense rewards and conservative scaling to improve Text-to-SQL performance. Third, we evaluate the impact of cold start, showing that distillation does not always benefit RLVR performance, and supervised fine-tuned models are prone to distributional mimicry. Fourth, we map the Pareto frontier of training efficiency, providing insights for optimizing Text-to-SQL reasoning under computational constraints. Our findings culminate in the Think2SQL family: our 4B-parameter model demonstrates reasoning capabilities competitive with state-of-the-art models such as o3. We release our models, datasets, and code to create a blueprint for RLVR optimization in Text-to-SQL at https://github.com/spapicchio/Think2SQL.
Show more
View full details
Poster

On Rate-Optimal Partitioning Classification from Observable and from Privatised Data

Balázs Cs. Csáji ⋅ László Györfi ⋅ Ambrus Tamás ⋅ Harro Walk
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P106
In this paper we revisit the classical method of partitioning classification and prove novel convergence rates under relaxed conditions, both for observable (non-privatised) and for privatised data. We consider the problem of classification in a d dimensional Euclidean space. Previous results on the partitioning classifier worked with the strong density assumption (SDA), which is restrictive, as we demonstrate through simple examples. Here, we study the problem under much milder assumptions. We presuppose that the distribution of the inputs is a mixture of an absolutely continuous and a discrete distribution, such that the absolutely continuous component is concentrated on a d_a dimensional subspace. In addition to the standard Lipschitz and margin conditions, a novel characteristic of the absolutely continuous component is introduced, by which the convergence rate of the classification error probability is computed, both for the binary and for the multi-class cases. This bound can reach the minimax optimal convergence rate achievable using SDA, but under much milder distributional assumptions. Interestingly, this convergence rate depends only on the intrinsic dimension of the continuous inputs, d_a, and not on d. Under privacy constraints, the data cannot be directly observed, and the constructed classifiers are functions of the randomised outcome of a suitable local differential privacy mechanism. In this paper we add Laplace distributed noises to the discretisations of all possible locations of the feature vector and to its label. Again, tight upper bounds on the convergence rate of the classification error probability can be derived, without using SDA, such that this rate depends on 2d_a.
Show more
View full details
Poster

Corruptions of Supervised Learning Problems: Typology and Mitigations

Laura Iacovissi ⋅ Nan Lu ⋅ Robert Williamson
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P107
Corruption is notoriously widespread in data collection. Despite extensive research, the existing literature predominantly focuses on specific settings and learning scenarios, lacking a unified view of corruption modelization and mitigation. In this work, we develop a general theory of corruption, which incorporates all modifications to a supervised learning problem, including changes in model class and loss. Focusing on changes to the underlying probability distributions via Markov kernels, our approach leads to three novel opportunities. First, it enables the construction of a novel, provably exhaustive corruption framework, distinguishing among different corruption types. This serves to unify existing models and establish a consistent nomenclature. Second, it facilitates a systematic analysis of corruption consequences on learning tasks, by considering Bayes risks in the clean and corrupted scenarios. Notably, while label corruptions affect only the loss function, attribute corruptions additionally influence the hypothesis class. Third, building upon these results, we investigate mitigations for various corruption types. We expand existing loss-correction methods for label corruption to handle dependent corruption types. Our findings highlight the necessity to generalize this classical corruption-corrected learning framework to a new paradigm with weaker requirements to encompass more corruption types. We provide such a paradigm as well as loss correction formulas in the attribute and joint corruption cases.
Show more
View full details
Poster

Autofocus Retrieval: An Effective Pipeline for Multi-Hop Question Answering With Semi-Structured Knowledge

Derian Boer ⋅ Stephen Roth ⋅ Stefan Kramer
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P301
In many real-world settings, machine learning models and interactive systems have access to both structured knowledge, e.g., knowledge graphs or tables, and unstructured content, e.g., natural language documents. Yet, most rely on either. Semi-Structured Knowledge Bases (SKBs) bridge this gap by linking unstructured content to nodes within structured data. In this work, we present Autofocus-Retriever (AF-Retriever), a modular framework for SKB-based, multi-hop question answering. It combines structural and textual retrieval through novel integration steps and optimizations, achieving the best zero- and one-shot results across all three STaRK QA benchmarks, which span diverse domains and evaluation metrics. AF-Retriever’s average first-hit rate surpasses the second-best method by 32.1%. Its performance is driven by (1) leveraging exchangeable large language models (LLMs) to extract entity attributes and relational constraints for both parsing and reranking the top- answers, (2) vector similarity search for ranking both extracted entities and final answers, (3) a novel incremental scope expansion procedure that prepares for the reranking on a configurable amount of suitable candidates that fulfill the given constraints the most, and (4) a hybrid retrieval strategy that reduces error susceptibility. In summary, while constantly adjusting the focus like an optical autofocus, AF-Retriever delivers a configurable amount of answer candidates in four constraint-driven retrieval steps, which are then supplemented and ranked through four additional processing steps. An ablation study and a detailed error analysis, including a comparison of three different LLM reranking strategies, provide component-level insights that are valuable for advancing the model and for enabling researchers and users to adapt, optimize, or extend its parts. The source code is publicly available at https://github.com/kramerlab/AF-Retriever.
Show more
View full details
Poster

Neural Fourier Transform for Multiple Time Series Prediction

Noam Koren ⋅ Daniel Freedman ⋅ Kira Radinsky
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P189
Multivariate time series forecasting is an important task in various fields such as economic planning, healthcare management, and environmental monitoring. In this work, we present a novel methodology for improving multivariate forecasting, particularly, in data sets with strong seasonality. We frame the forecasting task as a Multi-Dimensional Fourier Transform (MFT) problem and propose the Neural Fourier Transform (NFT) that leverages a deep learning model to predict future time series values by learning the MFT coefficients. The efficacy of NFT is empirically validated on 7 diverse datasets, demonstrating improvements over multiple forecasting horizons and lookbacks, thereby establishing new state-of-the-art results. Our contributions advance the field of multivariate time series forecasting by providing a model that excels in predictive accuracy.
Show more
View full details
Poster

FusionProt: Fusing Sequence and Structural Information for Unified Protein Representation Learning

Dan Kalifa ⋅ Uriel Singer ⋅ Kira Radinsky
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P305
Accurate protein representations that integrate sequence and three-dimensional (3D) structure are critical to many biological and biomedical tasks. Most existing models either ignore structure or combine it with sequence through a single, static fusion step. Here we present FusionProt, a unified model that learns representations via iterative, bidirectional fusion between a protein language model and a structure encoder. A single learnable token serves as a carrier, alternating between sequence attention and spatial message passing across layers. FusionProt is evaluated on Enzyme Commission (EC), Gene Ontology (GO), and mutation stability prediction tasks. It improves Fmax by a median of 1.3 points (up to 2.0) across EC and GO benchmarks, and boosts AUROC by 3.6 points over the strongest baseline on mutation stability. Inference cost remains practical, with only ~2-5% runtime overhead. Beyond state-of-the-art performance, we further demonstrate FusionProt’s practical relevance through representative biological case studies, suggesting that the model captures biologically relevant features.
Show more
View full details
Poster

On the Expressive Power and Limitations of Multi-Layer SSMs

Nikola Zubic ⋅ Qian Li ⋅ Yuyi Wang ⋅ Davide Scaramuzza
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P112
We study how depth, finite precision, state dimension, and chain-of-thought (CoT) affect the expressive power of multi-layer state-space models (SSMs). For the explicit-table $K$-function-composition problem, a canonical benchmark for sequential information propagation, we prove that any $L$-layer SSM solving $(L+3)$-function composition must satisfy $d^2p=\Omega(N/L^3)$, where $d$ is the state dimension and $p$ is the per-scalar precision. Conversely, $K$-function composition is solved exactly by a $(K+1)$-layer generalized SSM with $d=1$ and $p=\Theta(\log N)$. This gives a worst-case depth hierarchy for this formal problem family. We then distinguish post-input reasoning, in which all thought tokens are generated after the input, from input-interleaved reasoning, in which thought tokens may be inserted while the input stream is being read. Post-input reasoning does not circumvent our communication-based lower-bound pipeline, whereas input-interleaved reasoning admits bidirectional simulations with general deterministic one-pass streaming algorithms at the granularity of persistent memory. Finally, width and precision are not interchangeable under exact step-preserving simulation in the base affine-state model, but become interchangeable through the streaming-memory characterization once input-interleaved reasoning is allowed.
Show more
View full details
Poster

Beyond the Linear Separability Ceiling: Aligning Representations in VLMs

Enrico Vompa ⋅ Tanel Tammet ⋅ Mohit Vaishnav
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P268
A challenge in advancing Visual-Language Models (VLMs) is determining whether their failures on abstract reasoning tasks, such as Bongard problems, stem from flawed perception or faulty top-down reasoning. To disentangle these factors, we introduce a diagnostic framework centered on the Linear Separability Ceiling (LSC), the performance achievable by a linear classifier on a VLM's raw visual embeddings. Applying this framework to state-of-the-art VLMs, we uncover a pervasive ''alignment gap'', where most models fail to generatively outperform the linear separability of their representations. We find that the few models surpassing this ceiling do so via two mechanisms: by further refining visual representations into a more linearly separable format or by executing non-linear decision logic. We demonstrate that this bottleneck is not a fundamental limitation but a solvable visual alignment issue. Our method augments standard next-token prediction with a contrastive objective to restructure the visual manifold into a more one-dimensionally linear geometry, improving image-to-image comparison and enabling models to significantly surpass the LSC on abstract compositional reasoning tasks.
Show more
View full details
Poster

Online Bernstein-von Mises theorem

Jeyong Lee ⋅ Junhyeok Choi ⋅ Minwoo Chae
Online learning is an inferential paradigm in which parameters are updated incrementally from sequentially available data, in contrast to batch learning, where the entire dataset is processed at once. In this paper, we assume that mini-batches from the full dataset become available sequentially. The Bayesian framework, which updates beliefs about unknown parameters after observing each mini-batch, is naturally suited for online learning. At each step, we update the posterior distribution using the current prior and new observations, with the updated posterior serving as the prior for the next step. However, this recursive Bayesian updating is rarely computationally tractable unless the model and prior are conjugate. When the model is regular, the updated posterior can be approximated by a normal distribution, as justified by the Bernstein–von Mises theorem. We adopt a variational approximation at each step and investigate the frequentist properties of the final posterior obtained through this sequential procedure. Under mild assumptions, we show that the accumulated approximation error becomes negligible once the mini-batch size exceeds a threshold depending on the parameter dimension. As a result, the sequentially updated posterior is asymptotically indistinguishable from the full posterior.
Show more
View full details