Blog
Disaggregated Deep Learning via In-Physics Computing at Radio Frequency
Modern edge devices, such as cameras, drones, and Internet-of-Things nodes, rely on deep learning to enable a wide range of intelligent applications, including object recognition, environment perception, and autonomous navigation. Tuttavia, deploying deep learning models directly on the often resource-constrained edge devices...
On canonical differential equations for Calabi-Yau multi-scale Feynman integrals
We generalise a method recently introduced in the literature, that derives canonical differential equations, to multi-scale Feynman integrals with an underlying Calabi-Yau geometry. We start by recomputing a canonical form for the sunrise integral with all unequal masses. Inoltre, we compute for...
On the Degree Automatability of Sum-of-Squares Proofs
The Sum-of-Squares (SoS) hierarchy, also known as Lasserre hierarchy, has emerged as a promising tool in optimization. Tuttavia, it remains unclear whether fixed-degree SoS proofs can be automated [O’Donnell (2017)]. Indeed, there are examples of polynomial systems with bounded coefficients that admit...
First study of neutrino angle reconstruction using quasielastic-like interactions in MicroBooNE
We investigate the expected precision of the reconstructed neutrino direction using a {\nu}{\In}-argon quasielastic-like event topology with one muon and one proton in the final state and the reconstruction capabilities of the MicroBooNE liquid argon time projection chamber. This direction is of...
Bringing light into the Landau-Lifshitz-Gilbert equation: Consequences of its fractal non-Markovian memory kernel for optically induced magnetic inertia and magnons
The Landau-Lisfhitz-Gilbert (LLG) equation has been the cornerstone of modeling the dynamics of localized spins, viewed as classical vectors of fixed length, within nonequilibrium magnets. When light is employed as the nonequilibrium drive, the LLG equation must be supplemented with additional terms...
Zeptosecond free-electron compression through temporal lensing
The pursuit of ever-shorter time scales is a frontier in modern physics, exemplified by the synthesis of attosecond light pulses — an achievement made possible by coherently superimposing a broad range of photon energies, as required by the uncertainty principle. Tuttavia, extending...
Fitting Tree Metrics and Ultrametrics in Data Streams
Fitting distances to tree metrics and ultrametrics are two widely used methods in hierarchical clustering, primarily explored within the context of numerical taxonomy. Given a positive distance function $D:\binom{V}{2}\rightarrow\mathbb{R}_{>0}$, the goal is to find a tree (or ultrametric) $T$ including all elements...
Flexoelectric polarization in chiral liquid crystals: electrostatic self-interactions of topological defects
The presence of topological defects in apolar chiral liquid crystals cause orientational distortions, leading to non-uniform strain. This non-uniform strain generates an electric polarization response due to the flexoelectric effect, which induces an internal electric field. Associated to this electric field is...
Replay to Remember: Retaining Domain Knowledge in Streaming Language Models
Continual learning in large language models (LLM) typically encounters the critical challenge of catastrophic forgetting, where previously acquired knowledge deteriorates upon exposure to new data. While techniques like replay buffers and parameter-efficient tuning (per esempio., Low-Rank Adaptation or LoRA) have been proposed, few...
Token-Shuffle: Towards High-Resolution Image Generation with Autoregressive Models
Autoregressive (AR) modelli, long dominant in language generation, are increasingly applied to image synthesis but are often considered less competitive than Diffusion-based models. A primary limitation is the substantial number of image tokens required for AR models, which constrains both training and...