Paper: https://arxiv.org/abs/2602.03839 We built a system that enables distributed RL training over commodity internet connections. Weight synchronization drops from 14 GB to approximately 108 MB per update for a 7B model, completely lossless. Distributed RL separates training from inference. Training nodes remain centralized with fast interconnects, but inference nodes need fresh weights delivered over whatever network they have. For large models, this weight transfer becomes the bottleneck. Transferring 14 GB every few steps over commodity internet means waiting, […]

Ver mais

Trust shouldn’t be trapped inside apps. When you switch Nostr clients, your context disappears and you rebuild the same trust layer from scratch. I built an extension and oracle that compute Web of Trust scores from your social graph once, making them portable across every client. Users control trust, not apps.

Ver mais

arXiv:2603.08863v1 Announce Type: new Abstract: The stability and control of Unmanned Aerial Vehicles (UAVs) in a turbulent environment is a matter of great concern. Devising a robust control algorithm to reject disturbances is challenging due to the highly nonlinear nature of wind dynamics, and modeling the dynamics using analytical techniques is not straightforward. While traditional techniques using disturbance observers and classical adaptive control have shown some progress, they are mostly limited to relatively non-complex environments. On the other […]

Ver mais

arXiv:2601.21057v1 Announce Type: new Abstract: The recent success of deep learning (DL) has enabled the generation of high-quality synthetic gaze data. However, such data also raises privacy concerns because gaze sequences can encode subjects’ internal states, like fatigue, emotional load, or stress. Ideally, synthetic gaze should preserve the signal quality of real recordings and remove or attenuate state-related, privacy-sensitive attributes. Many recent DL-based generative models focus on replicating real gaze trajectories and do not explicitly consider subjective reports […]

Ver mais

arXiv:2602.07915v1 Announce Type: cross Abstract: Causal discovery from time series is a fundamental task in machine learning. However, its widespread adoption is hindered by a reliance on untestable causal assumptions and by the lack of robustness-oriented evaluation in existing benchmarks. To address these challenges, we propose CausalCompass, a flexible and extensible benchmark suite designed to assess the robustness of time-series causal discovery (TSCD) methods under violations of modeling assumptions. To demonstrate the practical utility of CausalCompass, we conduct […]

Ver mais

In the recent post, I got a lot of negative feedback for defending importance of Jesus being with you in Science (in the comment section). Do I feel wounded by that, not too much. What happened is different. From the early age I wanted that there is no secrets in the world, I believed whatever I felt had to be transmitted to the society. Why is that, because I believed when we hide something, it gives place to […]

Ver mais

arXiv:2505.03717v2 Announce Type: replace-cross Abstract: Low-rank matrix recovery is well-known to exhibit benign nonconvexity under the restricted isometry property (RIP): every second-order critical point is globally optimal, so local methods provably recover the ground truth. Motivated by the strong empirical performance of projected gradient methods for nonnegative low-rank recovery problems, we investigate whether this benign geometry persists when the factor matrices are constrained to be elementwise nonnegative. In the simple setting of a rank-1 nonnegative ground truth, we […]

Ver mais

arXiv:2602.12436v1 Announce Type: new Abstract: Barrier certificates, a form of state invariants, provide an automated approach to the verification of the safety of dynamical systems. Similarly to barrier certificates, recent works explore the notion of closure certificates, a form of transition invariants, to verify dynamical systems against $omega$-regular properties including safety. A closure certificate, defined over state pairs of a dynamical system, is a real-valued function whose zero superlevel set characterizes an inductive transition invariant of the system. […]

Ver mais

arXiv:2601.16302v1 Announce Type: new Abstract: Federated learning enables collaborative model training across geographically distributed medical centers while preserving data privacy. However, domain shifts and heterogeneity in data often lead to a degradation in model performance. Medical imaging applications are particularly affected by variations in acquisition protocols, scanner types, and patient populations. To address these issues, we introduce Federated Template and Task Learning (FeTTL), a novel framework designed to harmonize multi-institutional medical imaging data in federated environments. FeTTL learns […]

Ver mais

arXiv:2510.13907v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are highly sensitive to prompts, but most automatic prompt optimization (APO) methods assume access to ground-truth references (e.g., labeled validation data) that are costly to obtain. We propose the Prompt Duel Optimizer (PDO), a sample-efficient framework for label-free prompt optimization based on pairwise preference feedback from an LLM judge. PDO casts prompt selection as a dueling-bandit problem and combines (i) Double Thompson Sampling to prioritize informative comparisons under a […]

Ver mais