Neural Information Processing Systems http://nips.cc/

Event cameras are neuromorphic sensors that summarize the evolving world as a stream of events . Each event describes the pixel coordinates, time, and polarity of an intensity change.

We consider the problem of prediction with expert advice for ``easy'' sequences. We show that a variant of NormalHedge enjoys a second-order $ε$-quantile regret bound of $O\big(\sqrt{V_T \log(V_T/ε)}\big) $ when $V_T > \log N$, where $V_T$ is the cumulative second moment of instantaneous per-expert regret averaged with respect to a natural distribution determined by the algorithm. The algorithm is motivated by a continuous time limit using Stochastic Differential Equations. The discrete time analysis uses self-concordance techniques.

Weprovidestrong empirical evidence thatexactinference does not have this pathology, hence it is due to the approximation and not the model.

Isitpossibleto acquire therequired domain expertise, without involving human experts?

Wireframe parsing aims to recover line segments and their junctions to form a structured geometric representation useful for downstream tasks such as Simultaneous Localization and Mapping (SLAM). Existing methods predict lines and junctions separately and reconcile them post-hoc, causing mismatches and reduced robustness. We present Co-PLNet, a point-line collaborative framework that exchanges spatial cues between the two tasks, where early detections are converted into spatial prompts via a Point-Line Prompt Encoder (PLP-Encoder), which encodes geometric attributes into compact and spatially aligned maps. A Cross-Guidance Line Decoder (CGL-Decoder) then refines predictions with sparse attention conditioned on complementary prompts, enforcing point-line consistency and efficiency. Experiments on Wireframe and YorkUrban show consistent improvements in accuracy and robustness, together with favorable real-time efficiency, demonstrating our effectiveness for structured geometry perception.

We present a unified approach for constraint displacement problems in which a robot finds a feasible path by displacing constraints or obstacles. To this end, we propose a two stage process that returns locally optimal obstacle displacements to enable a feasible path for the robot. In the second stage, these obstacles are displaced to make the computed robot trajectory feasible, that is, collision-free. Several examples are provided that successfully demonstrate our approach on two distinct classes of constraint displacement problems. Introduction As humans, we encounter various situations in our day to day life in which we alter the location of objects - opening closed doors, repositioning chairs or other movable objects, clear objects while picking an object of interest from a cluttered table-top. As opposed to avoiding each object, altering or displacing these objects or constraints allow us to expand the solution space of feasible paths. In such situations, constraints, such as movable obstacles, may be cleared to find feasible paths. Manipulators often need to rearrange or move obstacles aside to accomplish a given set of tasks - a futuristic robot cooking dinner at home, manipulation in industrial settings, shelves replenishment in a grocery store. Service robots may need to reposition chairs or other movable objects to accomplish a task. A robot may need to plan a path through dynamic obstacles as they might clear the path while moving. We define a constraint displacement problem as one that finds a feasible path by displacing constraints while minimizing a problem-specific objective function.

The multi-armed bandit problem is a classic framework for sequential decision-making under uncertainty, encapsulating the fundamental trade-off between exploration and exploitation. This problem is typically studied in two primary settings: the stochastic setting, where the loss for each arm is drawn independently and identically from a fixed distribution, and the adversarial setting, where an arbitrary and potentially adaptive sequence of losses is generated. While numerous algorithms have been developed that are optimal for one of these settings, a significant challenge has been to design a single algorithm that performs well in both, without prior knowledge of the environment. Such an algorithm is said to have a "best-of-both-worlds" guarantee. By choosing an appropriate regularization function, a Follow-The-Regularized-Leader (FTRL) [Gordon, 1999a,b, Shalev-Shwartz and Singer, 2006a,b, Abernethy et al., 2008, Hazan and Kale, 2008] approach can achieve low regret in the adversarial setting while adapting to the easier stochastic setting to attain near-optimal, logarithmically-scaling regret, that is, a best-of-both-worlds guarantee. The Tsallis-INF algorithm [Audibert and Bubeck, 2009], when used in a FTRL algorithm was shown to have this property by Zimmert and Seldin [2021]. The aim of this note is to provide a concise and self-contained proof of this guarantee for the Tsallis-INF algorithm. In particular, the derivation relies on standard and modern techniques from online convex optimization, particularly the FTRL regret lemma and its local norm analysis.

Large Reasoning Models (LRMs) excel at complex reasoning but are traditionally evaluated in static, "frozen world" settings: model responses are assumed to be instantaneous, and the context of a request is presumed to be immutable over the duration of the response. While generally true for short-term tasks, the "frozen world" assumption breaks down in modern reasoning tasks such as assistive programming, where models may take hours to think through problems and code may change dramatically from the time the model starts thinking to the model's final output. In this work, we challenge the frozen world assumption and evaluate LRM robustness under two realistic dynamic scenarios: interruptions, which test the quality of the model's partial outputs on a limited budget, and dynamic context, which tests model adaptation to in-flight changes. Across mathematics and programming benchmarks that require long-form reasoning, static evaluations consistently overestimate robustness: even state-of-the-art LRMs, which achieve high accuracy in static settings, can fail unpredictably when interrupted or exposed to changing context, with performance dropping by up to 60% when updates are introduced late in the reasoning process. Our analysis further reveals several novel failure modes, including reasoning leakage, where models fold the reasoning into their final answer when interrupted; panic, where under time pressure models abandon reasoning entirely and return incorrect answers; and self-doubt, where performance degrades while incorporating updated information. Project Page: http://dynamic-lm.github.io/