arXiv subjects

Thomas Rossi

Publications and source records attributed to Thomas Rossi.

6 recordsLinked to original sources

arXiv2026

Uncertainty-gated selection for block-sparse attention

Block-sparse attention scales long-context language models by replacing the O(N^2) softmax with a per-query top-k selection over key blocks. This cutoff is myopic: when the k-th and (k+1)-th blocks are nearly tied in score, the selector commits without spending extra budget, and a dropped block carrying answer evidence is unrecoverable downstream. We propose a value-of-information router that measures, for each query, how decisively the top-k cut was made, and doubles the kept set for the queries where that gap is smallest; the rule is backbone-agnostic and stacks with existing block-scoring methods such as Quest. On LongBench-v2 medium at n=215 (the entire dataset subset), router-on-Quest reaches paired recall 0.75 vs. top-k 0.47 -- +28 pp over the SSA-style baseline (McNemar p<0.01) -- and lands within 2 pp of dense on RULER NIAH multikey at the same context. The lift reproduces on four models from three architectures (Qwen2.5, Mistral-Nemo, Qwen3.6). At 128K, the router preserves 0.81 and 0.89 of dense accuracy on Qwen2.5-7B-1M and Qwen3.6 (vs. SSA-style top-k at 0.09 on the former) while the fused selection-plus-kernel pipeline runs at 0.62x and 0.80x dense wall time.

cs.LG↗

arXiv2026

ActCam: Zero-Shot Joint Camera and 3D Motion Control for Video Generation

For artistic applications, video generation requires fine-grained control over both performance and cinematography, i.e., the actor's motion and the camera trajectory. We present ActCam, a zero-shot method for video generation that jointly transfers character motion from a driving video into a new scene and enables per-frame control of intrinsic and extrinsic camera parameters. ActCam builds on any pretrained image-to-video diffusion model that accepts conditioning in terms of scene depth and character pose. Given a source video with a moving character and a target camera motion, ActCam generates pose and depth conditions that remain geometrically consistent across frames. We then run a single sampling process with a two-phase conditioning schedule: early denoising steps condition on both pose and sparse depth to enforce scene structure, after which depth is dropped and pose-only guidance refines high-frequency details without over-constraining the generation. We evaluate ActCam on multiple benchmarks spanning diverse character motions and challenging viewpoint changes. We find that, compared to pose-only control and other pose and camera methods, ActCam improves camera adherence and motion fidelity, and is preferred in human evaluations, especially under large viewpoint changes. Our results highlight that careful camera-consistent conditioning and staged guidance can enable strong joint camera and motion control without training. Project page: https://elkhomar.github.io/actcam/.

cs.CV↗

arXiv2025

ReConForM : Real-time Contact-aware Motion Retargeting for more Diverse Character Morphologies

Preserving semantics, in particular in terms of contacts, is a key challenge when retargeting motion between characters of different morphologies. Our solution relies on a low-dimensional embedding of the character's mesh, based on rigged key vertices that are automatically transferred from the source to the target. Motion descriptors are extracted from the trajectories of these key vertices, providing an embedding that contains combined semantic information about both shape and pose. A novel, adaptive algorithm is then used to automatically select and weight the most relevant features over time, enabling us to efficiently optimize the target motion until it conforms to these constraints, so as to preserve the semantics of the source motion. Our solution allows extensions to several novel use-cases where morphology and mesh contacts were previously overlooked, such as multi-character retargeting and motion transfer on uneven terrains. As our results show, our method is able to achieve real-time retargeting onto a wide variety of characters. Extensive experiments and comparison with state-of-the-art methods using several relevant metrics demonstrate improved results, both in terms of motion smoothness and contact accuracy.

cs.GR↗

arXiv2023

Disentangling the Evolution of Electrons and Holes in photoexcited ZnO nanoparticles

The evolution of charge carriers in photoexcited room temperature ZnO nanoparticles in solution is investigated using ultrafast ultraviolet photoluminescence spectroscopy, ultrafast Zn K-edge absorption spectroscopy and ab-initio molecular dynamics (MD) simulations. The photoluminescence is excited at 4.66 eV, well above the band edge, and shows that electron cooling in the conduction band and exciton formation occur in <500 fs, in excellent agreement with theoretical predictions. The X-ray absorption measurements, obtained upon excitation close to the band edge at 3.49 eV, are sensitive to the migration and trapping of holes. They reveal that the 2 ps transient largely reproduces the previously reported transient obtained at 100 ps time delay in synchrotron studies. In addition, the X-ray absorption signal is found to rise in ~1.4 ps, which we attribute to the diffusion of holes through the lattice prior to their trapping at singly-charged oxygen vacancies. Indeed, the MD simulations show that impulsive trapping of holes induces an ultrafast expansion of the cage of Zn atoms in <200 fs, followed by an oscillatory response at a frequency of ~100 cm-1, which corresponds to a phonon mode of the system involving the Zn sub-lattice.

cond-mat.mtrl-sci↗

arXiv2021

Quantifying Photoinduced Polaronic Distortions in Inorganic Lead Halide Perovskites Nanocrystals

The development of next generation perovskite-based optoelectronic devices relies critically on the understanding of the interaction between charge carriers and the polar lattice in out-of-equilibrium conditions. While it has become increasingly evident for CsPbBr3 perovskites that the Pb-Br framework flexibility plays a key role in their light-activated functionality, the corresponding local structural rearrangement has not yet been unambiguously identified. In this work, we demonstrate that the photoinduced lattice changes in the system are due to a specific polaronic distortion, associated with the activation of a longitudinal optical phonon mode at 18 meV by electron-phonon coupling, and we quantify the associated structural changes with atomic-level precision. Key to this achievement is the combination of time-resolved and temperature-dependent studies at Br K-edge and Pb L3-edge X-ray absorption with refined ab-initio simulations, which fully account for the screened core-hole final state effects on the X-ray absorption spectra. From the temporal kinetics, we show that carrier recombination reversibly unlocks the structural deformation at both Br and Pb sites. The comparison with the temperature-dependent XAS results rules out thermal effects as the primary source of distortion of the Pb-Br bonding motif during photoexcitation. Our work provides a comprehensive description of the CsPbBr3 perovskites photophysics, offering novel insights on the light-induced response of the system and its exceptional optoelectronic properties.

cond-mat.mtrl-sci↗

arXiv2016

Localized holes and delocalized electrons in photoexcited inorganic perovskites: Watching each atomic actor by picosecond X-ray absorption spectroscopy

We report on an element-selective study of the fate of charge carriers in photoexcited inorganic CsPbBr3 and CsPb(ClBr)3 perovskite nanocrystals (NCs) in toluene solutions using time-resolved X-ray absorption spectroscopy with 80 ps time resolution. Probing the Br K-edge, the Pb L3-edge and the Cs L2-edge, we find that holes in the valence band are localized at Br atoms, forming small polarons, while electrons appear as delocalized in the conduction band. No signature of either electronic or structural changes are observed at the Cs L2-edge. The results at the Br and Pb edges suggest the existence of a weakly localized exciton, while the absence of signatures at the Cs edge indicates that the Cs+ cation plays no role in the charge transport, at least beyond 80 ps. These results can explain the rather modest charge carrier mobilities in these materials.

cond-mat.mtrl-sci↗