Publications of the month - April 2026

Premier Conference & Exhibition on Computer Graphics & Interactive Techniques (SIGGRAPH 2026), July 2026, Los Angeles

• Learning-based Sparse Signed Distance Field Super-Resolution by Sagar Panwar, Nissim Maruani (3IA Ph.D. student Alumni), Céline Loscos, Mathieu Desbrun, Pierre Alliez (3IA Chairholder)

Abstract: Signed Distance Fields (SDFs) are a powerful volumetric representation for 3D geometry. Recent advances in surface generation from SDFs increasingly rely on learnable surface representations and direct supervision on meshes. In this work, we challenge this trend and show that high-quality surface reconstruction can instead be achieved by learning to refine the volumetric signal itself. We present SuperSDF, a learning-based approach for sparse SDF super-resolution that operates directly in SDF space, without introducing any auxiliary surface representation or mesh-level supervision. Using a sparse voxel neural network restricted to a narrow band near the surface, our method predicts high-resolution signed-distance values from coarse inputs in a scalable and resolution-agnostic manner. Standard isosurface extraction algorithms can then process the resulting super-resolved SDFs, yielding accurate and detailed surface meshes. Our results demonstrate that learning-based SDF upsampling alone is sufficient to recover fine geometric details that are missed by classical interpolation and prior reconstruction methods. Compared to state-of-the-art ML approaches, our method produces higher-fidelity surfaces at a fraction of the computational cost and scales to volumetric resolutions previously out of reach.

Read the paper

64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), July 2026, San Diego (2026)

• Parallel Context-of-Experts Decoding for Retrieval Augmented Generation by Giulio Corallo, Paolo Papotti (3IA Chairholder)

Abstract: Retrieval Augmented Generation faces a trade-off: concatenating documents in a long prompt enables multi-document reasoning but creates prefill bottlenecks, while encoding document KV caches separately offers speed but breaks cross-document interaction. We propose Parallel Context-of-Experts Decoding (Pced), a training-free framework that shifts evidence aggregation from the attention mechanism to the decoding. Pced treats retrieved documents as isolated "experts", synchronizing their predictions via a novel retrieval-aware contrastive decoding rule that weighs expert logits against the model prior. This approach recovers cross-document reasoning capabilities without constructing a shared attention across documents.

Read the paper

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026), June 2026, Denver (USA)

• 3D Gaussian Splatting at Arbitrary Resolution with Compact Proxy Anchors by Mingyun Jeong, Seongro Yoon (3IA Ph.D. student), François Brémond (3IA Chairholder), Donghyeon Cho

Asbtract: Despite achieving high-quality rendering, 3D Gaussian Splatting suffers from aliasing when the rendering resolution changes, as it is typically trained at a fixed resolution. To address this limitation, we introduce a method that enables the model to generate resolution-adaptive 3D Gaussians under arbitrary resolution changes. In particular, built upon Scaffold-GS, we enhance the anchor feature representation by incorporating a resolution-embedding to encode continuous resolution information. From these enhanced anchor features, a pixel coverage gate dynamically forms resolution-adaptive 3D Gaussians. Furthermore, we drastically reduce storage requirements by selecting a compact subset of proxy anchors and designing a residual anchor predictor that reconstructs the unselected leaf anchors based on the proxy anchors, enabling faithful scene representation without compromising visual fidelity. As a result, our method provides continuous and alias-free rendering across resolutions while maintaining practical scalability and memory efficiency. Extensive experiments across diverse resolution ranges demonstrate that our approach achieves an optimal balance between fidelity and memory, enabling practical arbitrary resolution view synthesis even in resource-constrained settings.

Project page / Read the paper

29th Annual Conference on Artificial Intelligence and Statistics (AISTATS), May 2026, Tangier (Marocco)

• Rank Lifting and Random Non-Linear Maps by Andrea Drago, Maria Sofia Bucarelli (3IA Postdoctoral researcher), Francesco Caso, Marius Michetti, Federico Siciliano, Fabrizio Silvestri, Luca Becchetti

Abstract: Deep neural networks exhibit improved training and generalization performance as the number of parameters grows well beyond the size of the training set, contradicting classical intuitions about overfitting. In order to gain a better understanding of this “benign overparameterization”, we analyze the representational capacity of a random one-hidden-layer perceptron with Gaussian weights, no bias and threshold activations.More precisely, we investigate the following question: when does a hidden layer of dimension n maps k input vectors with pairwise angles at least $\theta$, to a full-rank activation matrix, thus ensuring that a simple linear classifier can perfectly fit those inputs in feature space? This problem has an immediate impact on memorization capacity at initialization and we frame it as a question about hyperplane arrangements on the unit sphere, and we prove new isoperimetric-like inequalities.  This allows us to derive non-trivial lower bounds on the probability that a random embedding avoids the arrangement’s zero-measure regions. Our results show that once the hidden dimension exceeds a threshold (depending on $\theta$ and the input dimension), hidden representations are linearly independent with high probability. While the case we consider is challenging due to the sparsity of the solution space, this setting highlights crucial, underlying geometric problems and connections to related questions in spherical geometry and linear algebra.

Read the paper

• The Majority Vote Paradigm Shift: When Popular Meets Optimal by Antonio Purificato, Maria Sofia Bucarelli (3IA Postdoctoral researcher), Anil Nelakanti, Andrea Bacciu, Fabrizio Silvestri, Amin Mantrach

Abstract: Reliably labelling data typically requires annotations from multiple human workers. However, humans are far from being perfect. Hence, it is a common practice to aggregate labels gathered from multiple annotators to make a more confident estimate of the true label. Among many aggregation methods, the simple and well-known Majority Vote (MV)  selects the class label receiving the highest number of votes. However, despite its importance, the optimality of MV’s label aggregation has not been extensively studied. We address this gap in our work by characterising the conditions under which MV achieves the theoretically optimal lower bound on label estimation error. Our results capture the tolerable limits on annotation noise under which MV can optimally recover labels for a given class distribution. This certificate of optimality provides a more principled approach to model selection for label aggregation as an alternative to otherwise inefficient practices that sometimes include higher experts, gold labels, etc., that are all marred by the same human uncertainty despite huge time and monetary costs. Experiments on both synthetic and real-world data corroborate our theoretical findings.

Read the paper

International Symposium on Biomedical Imaging (ISBI 2026), April 2026, London (England)

• Optimized High b-Value DWI Synthesis for Prostate Cancer Detection Using ADC Fusion by Fahym Bounazou, Francesco Boccarato, H L. Lee, Raphaele Renard Penna, Hervé Delingette (3IA Chairholder)

Abstract: High b-value diffusion-weighted imaging (DWI) enhances the visibility of prostate cancer lesions, but its acquisition often suffers from low signal-to-noise ratio (SNR) and significantly increases the patient scan times.
In this paper, we propose a method to synthesize high b-value DWI from standard multi-b DWI acquisitions using an optimized fusion of apparent diffusion coefficient (ADC) maps. A predicted ADC map is first estimated by linear regression from all available DWI volumes. When a scanner-provided ADC map is available, both maps are combined through a data-driven weighting strategy that accounts for the number of b-values used, the highest acquired b-value, and the contrast-to-noise ratio (CNR) in lesion regions. The fused ADC map is then used to extrapolate the signal at higher b-values.
Evaluations demonstrate that the proposed method preserves lesion contrast and anatomical details better than vendor-synthesized high b-value DWI, while remaining simple, interpretable, and easily applicable to multicentric datasets.

Read the paper

14th International Conference on Learning Representations (ICLR 2026), April 2026, Rio de Janeiro (Brazil)

• MASS: MoErging through Adaptive Subspace Selection by Donato Crisostomi, Alessandro Zirilli, Antonio Andrea Gargiulo, Maria Sofia Bucarelli (3IA Postdoctoral researcher), Simone Scardapane, Fabrizio Silvestri, Iacopo Masi, Emanuele Rodolà

Abstract: Model merging has recently emerged as a lightweight alternative to ensembling, combining multiple fine-tuned models into a single set of parameters with no additional training overhead. Yet, existing merging methods fall short of matching the full accuracy of separately fine-tuned endpoints. We present MASS (MoErging through Adaptive Subspace Selection), a new approach that closes this gap by unifying multiple fine-tuned models while retaining near state-of-the-art performance across tasks. Building on the low-rank decomposition of per-task updates, MASS stores only the most salient singular components for each task and merges them into a shared model. At inference time, a non-parametric, data-free router identifies which subspace (or combination thereof) best explains an input's intermediate features and activates the corresponding task-specific block. This procedure is fully training-free and introduces only a two-pass inference overhead plus a ~2 storage factor compared to a single pretrained model, irrespective of the number of tasks. We evaluate MASS on CLIP-based image classification using ViT-B-16, ViT-B-32 and ViT-L-14 for benchmarks of 8, 14 and 20 tasks respectively, establishing a new state-of-the-art. Most notably, MASS recovers up to ~98% of the average accuracy of individual fine-tuned models, making it a practical alternative to ensembling at a fraction of the storage cost.

Read the paper

Learning Meaningful Representations of Life (LMRL) Workshop @ ICLR 2026, April 2026, Rio de Janeiro (Brazil)

• AlphaSurf: On-the-Fly Surface Computations for Protein Representation Learning by Victor Gertner, Vincent Mallet, Frederic Cazals (3IA Chairholder)

Abstract: Several protein surfaces have been proposed for visualization purposes, and more. Recently, machine learning approaches incorporating surfaces as a biomolecular representation have emerged with strong performances, at the cost of increased computations. In this paper, we argue that this burden can be avoided. We propose a method to compute surfaces on-the-fly with no overhead and no performance loss.

Read the paper

Transactions on Machine Learning Research, October 2025

Improved seeding strategies for k-means and k-GMM by Guillaume Carrière, Frederic Cazals (3IA Chairholder)

Abstract: We revisit the randomized seeding techniques for k-means clustering and k-GMM (Gaussian Mixture model fitting with Expectation-Maximization), formalizing their three key ingredients: the metric used for seed sampling, the number of candidate seeds, and the metric used for seed selection. This analysis yields novel families of initialization methods exploiting a lookahead principle–conditioning the seed selection to an enhanced coherence with the final metric used to assess the algorithm, and a multipass strategy to tame down the effect of randomization. Experiments show a significant improvement over classical contenders. In particular, for k-means, our methods improve on the recently designed multi-swap strategy (similar results in terms of SSE, seeding ∼ ×6 faster), which was the first one to outperform the greedy k-means++ seeding. Our experimental analysis also shed light on subtle properties of k-means often overlooked, including the (lack of) correlations between the SSE upon seeding and the final SSE, the variance reduction phenomena observed in iterative seeding methods, and the sensitivity of the final SSE to the pool size for greedy methods. Practically, our most effective seeding methods are strong candidates to become one of the–if not the–standard techniques. From a theoretical perspective, our formalization of seeding opens the door to a new line of analytical approaches.

Read the paper