This paper introduces a learning-based control framework for a soft robotic actuator system designed to modulate intracranial pressure (ICP) waveforms, which is essential for studying cerebrospinal fluid dynamics and pathological processes underlying neurological disorders. A two-layer framework is proposed to safely achieve a desired ICP waveform modulation. First, a model predictive controller (MPC) with a disturbance observer is used for offset-free tracking of the system's motor position reference trajectory under safety constraints. Second, to address the unknown nonlinear dependence of ICP on the motor position, we employ a Bayesian optimization (BO) algorithm used for online learning of a motor position reference trajectory that yields the desired ICP modulation. The framework is experimentally validated using a test bench with a brain phantom that replicates realistic ICP dynamics in vitro. Compared to a previously employed proportional-integral-derivative controller, the MPC reduces mean and maximum motor position reference tracking errors by 83 % and 73 %, respectively. In less than 20 iterations, the BO algorithm learns a motor position reference trajectory that yields an ICP waveform with the desired mean and amplitude.

3D Gaussian Splatting (3DGS) has emerged as a dominant novel-view synthesis technique, but its high memory consumption severely limits its applicability on edge devices. A growing number of 3DGS compression methods have been proposed to make 3DGS more efficient, yet most only focus on storage compression and fail to address the critical bottleneck of rendering memory. To address this problem, we introduce MEGS$^{2}$, a novel memory-efficient framework that tackles this challenge by jointly optimizing two key factors: the total primitive number and the parameters per primitive, achieving unprecedented memory compression. Specifically, we replace the memory-intensive spherical harmonics with lightweight, arbitrarily oriented spherical Gaussian lobes as our color representations. More importantly, we propose a unified soft pruning framework that models primitive-number and lobe-number pruning as a single constrained optimization problem. Experiments show that MEGS$^{2}$ achieves a 50% static VRAM reduction and a 40% rendering VRAM reduction compared to existing methods, while maintaining comparable rendering quality. Project page: https://megs-2.github.io/

Deepfake detection remains a formidable challenge due to the complex and evolving nature of fake content in real-world scenarios. However, existing academic benchmarks suffer from severe discrepancies from industrial practice, typically featuring homogeneous training sources and low-quality testing images, which hinder the practical deployments of current detectors. To mitigate this gap, we introduce HydraFake, a dataset that simulates real-world challenges with hierarchical generalization testing. Specifically, HydraFake involves diversified deepfake techniques and in-the-wild forgeries, along with rigorous training and evaluation protocol, covering unseen model architectures, emerging forgery techniques and novel data domains. Building on this resource, we propose Veritas, a multi-modal large language model (MLLM) based deepfake detector. Different from vanilla chain-of-thought (CoT), we introduce pattern-aware reasoning that involves critical reasoning patterns such as "planning" and "self-reflection" to emulate human forensic process. We further propose a two-stage training pipeline to seamlessly internalize such deepfake reasoning capacities into current MLLMs. Experiments on HydraFake dataset reveal that although previous detectors show great generalization on cross-model scenarios, they fall short on unseen forgeries and data domains. Our Veritas achieves significant gains across different OOD scenarios, and is capable of delivering transparent and faithful detection outputs.

This paper presents a novel fabric-based thermal-haptic interface for virtual reality and teleoperation. It integrates pneumatic actuation and conductive fabric with an innovative ultra-lightweight design, achieving only 2~g for each finger unit. By embedding heating elements within textile pneumatic chambers, the system delivers modulated pressure and thermal stimuli to fingerpads through a fully soft, wearable interface. Comprehensive characterization demonstrates rapid thermal modulation with heating rates up to 3$^{\circ}$C/s, enabling dynamic thermal feedback for virtual or teleoperation interactions. The pneumatic subsystem generates forces up to 8.93~N at 50~kPa, while optimization of fingerpad-actuator clearance enhances cooling efficiency with minimal force reduction. Experimental validation conducted with two different user studies shows high temperature identification accuracy (0.98 overall) across three thermal levels, and significant manipulation improvements in a virtual pick-and-place tasks. Results show enhanced success rates (88.5\% to 96.4\%, p = 0.029) and improved force control precision (p = 0.013) when haptic feedback is enabled, validating the effectiveness of the integrated thermal-haptic approach for advanced human-machine interaction applications.

Probabilistic decoding in Large Language Models (LLMs) often yields inconsistent outputs, particularly on complex or long-form questions. Self-Consistency (SC) mitigates this for short-form QA by majority voting over exact strings, whereas Universal Self-Consistency (USC) and Weighted Unigram Consistency Score (WUCS) extend to long-form responses but lose accuracy on short-form benchmarks. We introduce \textbf{Latent Self-Consistency (LSC)}, which selects the most semantically consistent response using learnable token embeddings. LSC's lightweight forward processing of summary tokens only introduces negligible runtime overhead (at most $0.9\%$) on top of standard decoding of the base LLM, and requires no changes to the model architecture. Across 6 short-form and 5 long-form reasoning benchmarks (e.g., MATH, MMLU, TruthfulQA), LSC surpasses SC, USC, and WUCS on both short-form and long-form on average performance, while adding negligible computational overhead on vanilla inference. These results position LSC as a reliable consistency-selection method that works effectively across various answer formats. Additionally, LSC provides well-calibrated confidence estimates, maintaining low expected calibration error across both answer formats.

Existing reinforcement learning (RL) methods struggle with long-horizon robotic manipulation tasks, particularly those involving sparse rewards. While action chunking is a promising paradigm for robotic manipulation, using RL to directly learn continuous action chunks in a stable and data-efficient manner remains a critical challenge. This paper introduces AC3 (Actor-Critic for Continuous Chunks), a novel RL framework that learns to generate high-dimensional, continuous action sequences. To make this learning process stable and data-efficient, AC3 incorporates targeted stabilization mechanisms for both the actor and the critic. First, to ensure reliable policy improvement, the actor is trained with an asymmetric update rule, learning exclusively from successful trajectories. Second, to enable effective value learning despite sparse rewards, the critic's update is stabilized using intra-chunk $n$-step returns and further enriched by a self-supervised module providing intrinsic rewards at anchor points aligned with each action chunk. We conducted extensive experiments on 25 tasks from the BiGym and RLBench benchmarks. Results show that by using only a few demonstrations and a simple model architecture, AC3 achieves superior success rates on most tasks, validating its effective design.

Real-world forecasting requires models to integrate not only historical data but also relevant contextual information provided in textual form. While large language models (LLMs) show promise for context-aided forecasting, critical challenges remain: we lack diagnostic tools to understand failure modes, performance remains far below their potential, and high computational costs limit practical deployment. We introduce a unified framework of four strategies that address these limitations along three orthogonal dimensions: model diagnostics, accuracy, and efficiency. Through extensive evaluation across model families from small open-source models to frontier models including Gemini, GPT, and Claude, we uncover both fundamental insights and practical solutions. Our findings span three key dimensions: diagnostic strategies reveal the "Execution Gap" where models correctly explain how context affects forecasts but fail to apply this reasoning; accuracy-focused strategies achieve substantial performance improvements of 25-50%; and efficiency-oriented approaches show that adaptive routing between small and large models can approach large model accuracy on average while significantly reducing inference costs. These orthogonal strategies can be flexibly integrated based on deployment constraints, providing practitioners with a comprehensive toolkit for practical LLM-based context-aided forecasting.

Generative models, especially diffusion and flow-based models, have been promising in offline multi-agent reinforcement learning. However, integrating powerful generative models into this framework poses unique challenges. In particular, diffusion and flow-based policies suffer from low sampling efficiency due to their iterative generation processes, making them impractical in time-sensitive or resource-constrained settings. To tackle these difficulties, we propose OM2P (Offline Multi-Agent Mean-Flow Policy), a novel offline MARL algorithm to achieve efficient one-step action sampling. To address the misalignment between generative objectives and reward maximization, we introduce a reward-aware optimization scheme that integrates a carefully-designed mean-flow matching loss with Q-function supervision. Additionally, we design a generalized timestep distribution and a derivative-free estimation strategy to reduce memory overhead and improve training stability. Empirical evaluations on Multi-Agent Particle and MuJoCo benchmarks demonstrate that OM2P achieves superior performance, with up to a 3.8x reduction in GPU memory usage and up to a 10.8x speed-up in training time. Our approach represents the first to successfully integrate mean-flow model into offline MARL, paving the way for practical and scalable generative policies in cooperative multi-agent settings.

Recent advances in 3D scene reconstruction and 4D human animation have broadened adoption, but integrating the two remains difficult. Key challenges include placing humans at plausible locations and scales without interpenetration, aligning lighting and style between humans and background, and handling dynamic camera trajectories. We introduce AnimateScene, a unified framework addressing these issues. First, an accurate placement module automatically determines realistic 3D positions and prevents collisions during motion. Second, we propose a training-free style alignment method that adapts the 4D human to match background illumination and appearance, achieving coherent composites. Finally, we design a joint post-reconstruction approach that incorporates camera trajectories for smooth, visually engaging motion videos. Experiments across diverse scenes and actions demonstrate that AnimateScene generates dynamic results with high geometric detail and strong spatiotemporal coherence. The project page is available at https://whynothaha.github.io/AnimateScene/AnimateScene.html.

Effective robotic systems for long-horizon human-robot collaboration must adapt to a wide range of human partners, whose physical behavior, willingness to assist, and understanding of the robot's capabilities may change over time. This demands a tightly coupled communication loop that grants both agents the flexibility to propose, accept, or decline requests as they coordinate toward completing the task effectively. We apply a Mixed-Initiative dialog paradigm to Collaborative human-roBot teaming and propose MICoBot, a system that handles the common scenario where both agents, using natural language, take initiative in formulating, accepting, or rejecting proposals on who can best complete different steps of a task. To handle diverse, task-directed dialog, and find successful collaborative strategies that minimize human effort, MICoBot makes decisions at three levels: (1) a meta-planner considers human dialog to formulate and code a high-level collaboration strategy, (2) a planner optimally allocates the remaining steps to either agent based on the robot's capabilities (measured by a simulation-pretrained affordance model) and the human's estimated availability to help, and (3) an action executor decides the low-level actions to perform or words to say to the human. In physical robot trials with 18 unique human participants, MICoBot significantly improves task success and user experience over a pure LLM baseline and standard agent allocation models. See additional videos and materials at https://robin-lab.cs.utexas.edu/MicoBot/.

We present specialized Large Language Models for theoretical High-Energy Physics, obtained as 20 fine-tuned variants of the 8-billion parameter Llama-3.1 model. Each variant was trained on arXiv abstracts (through August 2024) from different combinations of hep-th, hep-ph and gr-qc. For a comparative study, we also trained models on datasets that contained abstracts from disparate fields such as the q-bio and cs categories. All models were fine-tuned using two distinct Low-Rank Adaptation fine-tuning approaches and varying dataset sizes, and outperformed the base model on hep-th abstract completion tasks. We compare performance against leading commercial LLMs (ChatGPT, Claude, Gemini, DeepSeek) and derive insights for further developing specialized language models for High-Energy Theoretical Physics.

Modern approaches to enhancing Large Language Models' factual accuracy and knowledge utilization face a fundamental trade-off: non-parametric retrieval-augmented generation (RAG) provides flexible access to external knowledge but suffers from high inference latency and shallow integration, while parametric fine-tuning methods like LoRA risk catastrophic forgetting and degraded general capabilities. In this work, we propose MLP Memory, a lightweight parametric module that learns to internalize retrieval patterns without explicit document access. By pretraining an MLP to imitate a $k$NN retriever's behavior on the entire pretraining dataset, we create a differentiable memory component that captures the benefits of retrieval-based knowledge access in a fully parametric form. Our architecture integrates this pretrained MLP Memory with Transformer decoders through simple probability interpolation, yielding 17.5\% and 24.1\% scaling gains on WikiText-103 and Web datasets, respectively. It further achieves 12.3\% relative improvement on five question-answering benchmarks and 5.2 points absolute gain across nine general NLP tasks, while reducing hallucinations by up to 10 points on HaluEval. Moreover, MLP Memory delivers 2.5$\times$ faster inference than RAG with superior accuracy. Our findings show that learning retrieval patterns parametrically bridges the gap between efficient inference and effective knowledge access, offering a practical alternative to both RAG and fine-tuning approaches.

Adapting generative Multimodal Large Language Models (MLLMs) into universal embedding models typically demands resource-intensive contrastive pre-training, while traditional hard negative mining methods suffer from severe false negative contamination. In this paper, we propose a highly data-efficient framework that bypasses extensive pre-training to build a robust multimodal representation space. We first introduce a hierarchical embedding prompt that provides strong latent conditioning. By explicitly anchoring task definitions at the system level, this prompting strategy effectively bridges the modality gap and unlocks powerful zero-shot embedding capabilities. Building upon this latent conditioning, we present Self-aware Hard Negative Sampling (SaHa). Unlike conventional candidate-space mining, SaHa shifts the mechanism to the query-space by mapping retrieved candidates back to their owner queries to rigorously filter out semantic false negatives. Furthermore, our method constructs mutually hard clusters, maximizing intra-task discrimination and batch efficiency without redundant forward passes. Extensive experiments demonstrate that our unified approach achieves highly competitive fine-tuning performance on the Massive Multimodal Embedding Benchmark using only a fraction of standard training data.

Text-to-Image (T2I) models generate high-quality images but are vulnerable to malicious backdoor attacks that inject harmful biases (e.g., trigger-activated gender or racial stereotypes). Existing debiasing methods, often designed for natural statistical biases, struggle with these deliberately and subtly injected attacks. We propose AutoDebias, a framework that automatically identifies and mitigates these malicious biases in T2I models without prior knowledge of the specific attack types. Specifically, AutoDebias leverages vision-language models to detect trigger-activated visual patterns and constructs neutralization guides by generating counter-prompts. These guides drive a CLIP-guided training process that breaks the harmful associations while preserving the original model's image quality and diversity. Unlike methods designed for natural bias, AutoDebias effectively addresses subtle, injected stereotypes and multiple interacting attacks. We evaluate the framework on a new benchmark covering 17 distinct backdoor scenarios, including challenging cases where multiple backdoors co-exist. AutoDebias detects malicious patterns with 91.6% accuracy and reduces the backdoor success rate from 90% to negligible levels, while preserving the visual fidelity of the original model.

Efficient and high-accuracy 3D occupancy prediction is vital for the performance of autonomous driving systems. However, existing methods struggle to balance precision and efficiency: high-accuracy approaches are often hindered by heavy computational overhead, leading to slow inference speeds, while others leverage pure bird's-eye-view (BEV) representations to gain speed at the cost of losing vertical spatial cues and compromising geometric integrity. To overcome these limitations, we build on the efficient Lift-Splat-Shoot (LSS) paradigm and propose a pure 2D framework, DA-Occ, for 3D occupancy prediction that preserves fine-grained geometry. Standard LSS-based methods lift 2D features into 3D space solely based on depth scores, making it difficult to fully capture vertical structure. To improve upon this, DA-Occ augments depth-based lifting with a complementary height-score projection that explicitly encodes vertical geometric information. We further employ direction-aware convolution to extract geometric features along both vertical and horizontal orientations, effectively balancing accuracy and computational efficiency. On the Occ3D-nuScenes, the proposed method achieves an mIoU of 39.3% and an inference speed of 27.7 FPS, effectively balancing accuracy and efficiency. In simulations on edge devices, the inference speed reaches 14.8 FPS, further demonstrating the method's applicability for real-time deployment in resource-constrained environments.

Construction throughout history typically assumes that its blueprints and building blocks are pre-determined. However, recent work suggests that alternative approaches can enable new paradigms for structure formation. Aleatory architectures, or those which rely on the properties of their granular building blocks rather than pre-planned design or computation, have thus far relied on human intervention for their creation. We imagine that robotic swarms could be valuable to create such aleatory structures by manipulating and forming structures from entangled granular materials. To discover principles by which robotic systems can effectively manipulate soft matter, we develop a robophysical model for interaction with geometrically cohesive granular media composed of u-shape particles. This robotic platform uses environmental signals to autonomously coordinate excavation, transport, and deposition of material. We test the effect of substrate initial conditions by characterizing robot performance in two different material compaction states and observe as much as a 75% change in transported mass depending on initial substrate compressive loading. These discrepancies suggest the functional role that material properties such as packing and cohesion/entanglement play in excavation and construction. To better understand these material properties, we develop an apparatus for tensile testing of the geometrically cohesive substrates, which reveals how entangled material strength responds strongly to initial compressive loading. These results explain the variation observed in robotic performance and point to future directions for better understanding robotic interaction mechanics with entangled materials.

The advent of novel view synthesis techniques such as NeRF and 3D Gaussian Splatting (3DGS) has enabled learning precise 3D models only from posed monocular images. Although these methods are attractive, they hold two major limitations that prevent their use in space applications: they require poses during training, and have high computational cost at training and inference. To address these limitations, this work contributes: (1) a Convolutional Neural Network (CNN) based primitive initializer for 3DGS using monocular images; (2) a pipeline capable of training with noisy or implicit pose estimates; and (3) and analysis of initialization variants that reduce the training cost of precise 3D models. A CNN takes a single image as input and outputs a coarse 3D model represented as an assembly of primitives, along with the target's pose relative to the camera. This assembly of primitives is then used to initialize 3DGS, significantly reducing the number of training iterations and input images needed -- by at least an order of magnitude. For additional flexibility, the CNN component has multiple variants with different pose estimation techniques. This work performs a comparison between these variants, evaluating their effectiveness for downstream 3DGS training under noisy or implicit pose estimates. The results demonstrate that even with imperfect pose supervision, the pipeline is able to learn high-fidelity 3D representations, opening the door for the use of novel view synthesis in space applications.

This position paper examines the use of Large Language Models (LLMs) in social simulation, analyzing their potential and limitations from a computational social science perspective. We first review recent findings on LLMs' ability to replicate key aspects of human cognition, including Theory of Mind reasoning and social inference, while identifying persistent limitations such as cognitive biases, lack of grounded understanding, and behavioral inconsistencies. We then survey emerging applications of LLMs in multi-agent simulation frameworks, examining system architectures, scalability, and validation strategies. Projects such as Generative Agents (Smallville) and AgentSociety are analyzed with respect to their empirical grounding and methodological design. Particular attention is given to the challenges of behavioral fidelity, calibration, and reproducibility in large-scale LLM-driven simulations. Finally, we distinguish between contexts where LLM-based agents provide operational value-such as interactive simulations and serious games-and contexts where their use raises epistemic concerns, particularly in explanatory or predictive modeling. We argue that hybrid approaches integrating LLMs into established agent-based modeling platforms such as GAMA and NetLogo may offer a promising compromise between expressive flexibility and analytical transparency. Building on this analysis, we outline a conceptual research direction termed Hybrid Constitutional Architectures, which proposes a stratified integration of classical agent-based models (ABMs), small language models (SLMs), and LLMs within established platforms such as GAMA and NetLogo.

The dynamic allocation of spectrum in 5G / 6G networks is critical to efficient resource utilization. However, applying traditional deep reinforcement learning (DRL) is often infeasible due to its immense sample complexity and the safety risks associated with unguided exploration, which can cause severe network interference. To address these challenges, we propose a meta-learning framework that enables agents to learn a robust initial policy and rapidly adapt to new wireless scenarios with minimal data. We implement three meta-learning architectures, model-agnostic meta-learning (MAML), recurrent neural network (RNN), and an attention-enhanced RNN, and evaluate them against a non-meta-learning DRL algorithm, proximal policy optimization (PPO) baseline, in a simulated dynamic integrated access/backhaul (IAB) environment. Our results show a clear performance gap. The attention-based meta-learning agent reaches a peak mean network throughput of 48 Mbps, while the PPO baseline decreased drastically to 10 Mbps. Furthermore, our method reduces SINR and latency violations by more than 50% compared to PPO. It also shows quick adaptation, with a fairness index 0.7, showing better resource allocation. This work proves that meta-learning is a very effective and safer option for intelligent control in complex wireless systems.

Machine learning (ML) algorithms have emerged in many meteorological applications. However, these algorithms struggle to extrapolate beyond the data they were trained on, i.e., they may adopt faulty strategies that lead to catastrophic failures. These failures are difficult to predict due to the opaque nature of ML algorithms. In high-stakes applications, such as severe weather forecasting, is is crucial to avoid such failures. One approach to address this issue is to develop more interpretable ML algorithms. The primary goal of this work is to illustrate the use of a specific interpretable ML algorithm that has not yet found much use in meteorology, Explainable Boosting Machines (EBMs). We demonstrate that EBMs are particularly suitable to implement human-guided strategies in an ML algorithm. As guiding example, we show how to develop an EBM to detect overshooting tops (OTs) in satellite imagery. EBMs require input features to be scalar. We use techniques from Knowledge-Guided Machine Learning to first extract scalar features from meteorological imagery. For the application of identifying OTs this includes extracting cloud texture from satellite imagery using Gray-Level Co-occurrence Matrices. Once trained, the EBM was examined and minimally altered to more closely match strategies used by domain scientists to identify OTs. The result of our efforts is a fully interpretable ML algorithm developed in a human-machine collaboration that uses human-guided strategies. While the final model does not reach the accuracy of more complex approaches, it performs reasonably well and we hope paves the way for building more interpretable ML algorithms for this and other meteorological applications.

Small-scale Vision-Language Models (SVLMs) are exceptionally well-suited for proprietary tasks. Equipping them with thinking capabilities is a critical step to enhance their performance and reliability in these specific domains. However, existing training paradigms, including Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Reward (RLVR), impose substantial demands on the base VLM, exceeding the capacity of SVLMs. Consequently, directly applying these paradigms to SVLMs fails to instill the desired thinking abilities. A natural solution is to combine SFT and RLVR, leveraging their complementarity to reduce the dependence on model capacity. Yet the core challenge lies in managing the inherent trade-off: excessive reliance on SFT can force the model to memorize pseudo thinking traces, while over-emphasizing RLVR can lead to unstable exploration (i.e., advantage collapse). To address this, we propose DyME, a novel training paradigm that Dynamically selects between Memorization (via SFT) and Exploration (via RLVR) at each optimization step. By ensuring that every update contributes to the trade-off, DyME serves as a robust, standalone strategy that stabilizes SVLM learning. Complementing this paradigm, we further introduce a synergistic Visual Supervision mechanism (comprising a visual checker and refiner) designed to inject dynamically enhanced, image-grounded guidance during optimization. Extensive experiments across diverse domains demonstrate that DyME consistently achieves this balance, and thus delivers substantial performance improvements on specialized tasks. These results establish DyME as a practical and effective solution for empowering SVLMs with reliable thinking capabilities. GitHub: https://github.com/HKUST-LongGroup/DyME

Large language model (LLM) agents show promise in automating machine learning (ML) engineering. However, existing agents typically operate in isolation on a given research problem, without engaging with the broader research community, where human researchers often gain insights and contribute by sharing knowledge. To bridge this gap, we introduce MLE-Live, a live evaluation framework designed to assess an agent's ability to communicate with and leverage collective knowledge from a simulated Kaggle research community. Building on this framework, we propose CoMind, a multi-agent system designed to systematically leverage external knowledge. CoMind employs an iterative parallel exploration mechanism, developing multiple solutions simultaneously to balance exploratory breadth with implementation depth. On 75 past Kaggle competitions within our MLE-Live framework, CoMind achieves a 36% medal rate, establishing a new state of the art. Critically, when deployed in eight live, ongoing competitions, CoMind outperforms 92.6% of human competitors on average, placing in the top 5% on three official leaderboards and the top 1% on one.

Conventional knowledge distillation, designed for model compression, fails on long-tailed distributions because the teacher model tends to be biased toward head classes and provides limited supervision for tail classes. We propose Long-Tailed Knowledge Distillation (LTKD), a novel framework that reformulates the conventional objective into two components: a cross-group loss, capturing mismatches in prediction distributions across class groups (head, medium, and tail), and a within-group loss, capturing discrepancies within each group's distribution. This decomposition reveals the specific sources of the teacher's bias. To mitigate the inherited bias, LTKD introduces (1) a rebalanced cross-group loss that calibrates the teacher's group-level predictions and (2) a reweighted within-group loss that ensures equal contribution from all groups. Extensive experiments on CIFAR-100-LT, TinyImageNet-LT, and ImageNet-LT demonstrate that LTKD significantly outperforms existing methods in both overall and tail-class accuracy, thereby showing its ability to distill balanced knowledge from a biased teacher for real-world applications.

Recent advancements in 4D generation have demonstrated its remarkable capability in synthesizing photorealistic renderings of dynamic 3D scenes. However, despite achieving impressive visual performance, almost all existing methods overlook the generation of spatial audio aligned with the corresponding 4D scenes, posing a significant limitation to truly immersive audiovisual experiences. To mitigate this issue, we propose Sonic4D, a novel framework that enables spatial audio generation for immersive exploration of 4D scenes. Specifically, our method is composed of three stages: 1) To capture both the dynamic visual content and raw auditory information from a monocular video, we first employ pre-trained expert models to generate the 4D scene and its corresponding monaural audio. 2) Subsequently, to transform the monaural audio into spatial audio, we localize and track the sound sources within the 4D scene, where their 3D spatial coordinates at different timestamps are estimated via a pixel-level visual grounding strategy. 3) Based on the estimated sound source locations, we further synthesize plausible spatial audio that varies across different viewpoints and timestamps using physics-based simulation. Extensive experiments have demonstrated that our proposed method generates realistic spatial audio consistent with the synthesized 4D scene in a training-free manner, significantly enhancing the immersive experience for users. Generated audio and video examples are available at https://x-drunker.github.io/Sonic4D-project-page.

Human trajectory forecasting is crucial in applications such as autonomous driving, robotics and surveillance. Accurate forecasting requires models to consider various factors, including social interactions, multi-modal predictions, pedestrian intention and environmental context. While existing methods account for these factors, they often overlook the impact of the environment, which leads to collisions with obstacles. This paper introduces ECAM (Environmental Collision Avoidance Module), a contrastive learning-based module to enhance collision avoidance ability with the environment. The proposed module can be integrated into existing trajectory forecasting models, improving their ability to generate collision-free predictions. We evaluate our method on the ETH/UCY dataset and quantitatively and qualitatively demonstrate its collision avoidance capabilities. Our experiments show that state-of-the-art methods significantly reduce (-40/50%) the collision rate when integrated with the proposed module. The code is available at https://github.com/CVML-CFU/ECAM.

The use of target networks in deep reinforcement learning is a widely popular solution to mitigate the brittleness of semi-gradient approaches and stabilize learning. However, target networks notoriously require additional memory and delay the propagation of Bellman updates compared to an ideal target-free approach. In this work, we step out of the binary choice between target-free and target-based algorithms. We introduce a new method that uses a copy of the last linear layer of the online network as a target network, while sharing the remaining parameters with the up-to-date online network. This simple modification enables us to keep the target-free's low-memory footprint while leveraging the target-based literature. We find that combining our approach with the concept of iterated $Q$-learning, which consists of learning consecutive Bellman updates in parallel, helps improve the sample-efficiency of target-free approaches. Our proposed method, iterated Shared $Q$-Learning (iS-QL), bridges the performance gap between target-free and target-based approaches across various problems while using a single $Q$-network, thus stepping towards resource-efficient reinforcement learning algorithms.

While Large Language Models (LLMs) achieve near-human performance on standard benchmarks, their capabilities often fail to generalize to complex, real-world problems. To bridge this gap, we introduce DeepQuestion, a scalable, automated framework that systematically elevates the cognitive complexity of existing datasets. Grounded in Bloom's taxonomy, DeepQuestion generates (1) scenario-based problems to test the application of knowledge in noisy, realistic contexts, and (2) instruction-based prompts that require models to create new questions from a given solution path, assessing synthesis and evaluation skills. Our extensive evaluation across ten leading open-source and proprietary models reveals a stark performance decline with accuracy dropping by up to 70% as tasks ascend the cognitive hierarchy. These findings underscore that current benchmarks overestimate true reasoning abilities and highlight the critical need for cognitively diverse evaluations to guide future LLM development.

The Lipschitz constant of a neural network is connected to several important properties of the network such as its robustness and generalization. It is thus useful in many settings to estimate the Lipschitz constant of a model. Prior work has focused mainly on estimating the Lipschitz constant of multi-layer perceptrons and convolutional neural networks. Here we focus on data modeled as sets or multi-sets of vectors and on neural networks that can handle such data. These models typically apply some permutation invariant aggregation function, such as the sum, mean or max operator, to the input multisets to produce a single vector for each input sample. In this paper, we investigate whether these aggregation functions, along with an attention-based aggregation function, are Lipschitz continuous with respect to three distance functions for unordered multisets, and we compute their Lipschitz constants. In the general case, we find that each aggregation function is Lipschitz continuous with respect to only one of the three distance functions, while the attention-based function is not Lipschitz continuous with respect to any of them. Then, we build on these results to derive upper bounds on the Lipschitz constant of neural networks that can process multisets of vectors, while we also study their stability to perturbations and generalization under distribution shifts. To empirically verify our theoretical analysis, we conduct a series of experiments on datasets from different domains.

Image editing is an important task in computer graphics, vision, and VFX, with recent diffusion-based methods achieving fast and high-quality results. However, edits requiring significant structural changes, such as non-rigid deformations, object modifications, or content generation, remain challenging. Existing few step editing approaches produce artifacts such as irrelevant texture or struggle to preserve key attributes of the source image (e.g., pose). We introduce Cora, a novel editing framework that addresses these limitations by introducing correspondence-aware noise correction and interpolated attention maps. Our method aligns textures and structures between the source and target images through semantic correspondence, enabling accurate texture transfer while generating new content when necessary. Cora offers control over the balance between content generation and preservation. Extensive experiments demonstrate that, quantitatively and qualitatively, Cora excels in maintaining structure, textures, and identity across diverse edits, including pose changes, object addition, and texture refinements. User studies confirm that Cora delivers superior results, outperforming alternatives.

Knowledge editing methods (KEs) are a cost-effective way to update the factual content of large language models (LLMs), but they pose a dual-use risk. While KEs are beneficial for updating outdated or incorrect information, they can be exploited maliciously to implant misinformation or bias. In order to defend against these types of malicious manipulation, we need robust techniques that can reliably detect, interpret, and mitigate malicious edits. To that end, we introduce the tasks of tracing and reversing edits. We propose a novel method to infer the edited object entity, solely based on the modified weights, without access to the editing prompt or any other semantically similar prompts, with up to 99% accuracy. Further, we propose an effective and training-free method for reversing edits. Our method reverses up to 94% of the edits, and helps regain the original model's output distribution without access to any information about the edit. This method can further be repurposed to distinguish between edited and unedited weights. Our findings highlight the feasibility of tracing and reversing edits based on the edited weights, opening a new research direction for safeguarding LLMs against adversarial manipulations.