id
stringlengths
9
16
title
stringlengths
4
278
categories
stringlengths
5
104
abstract
stringlengths
6
4.09k
2502.11913
PreAdaptFWI: Pretrained-Based Adaptive Residual Learning for Full-Waveform Inversion Without Dataset Dependency
physics.geo-ph cs.LG
Full-waveform inversion (FWI) is a method that utilizes seismic data to invert the physical parameters of subsurface media by minimizing the difference between simulated and observed waveforms. Due to its ill-posed nature, FWI is susceptible to getting trapped in local minima. Consequently, various research efforts have attempted to combine neural networks with FWI to stabilize the inversion process. This study presents a simple yet effective training framework that is independent of dataset reliance and requires only moderate pre-training on a simple initial model to stabilize network outputs. During the transfer learning phase, the conventional FWI gradients will simultaneously update both the neural network and the proposed adaptive residual learning module, which learns the residual mapping of large-scale distribution features in the network's output, rather than directly fitting the target mapping. Through this synergistic training paradigm, the proposed algorithm effectively infers the physically-informed prior knowledge into a global representation of stratigraphic distribution, as well as capturing subtle variations in inter-layer velocities within local details, thereby escaping local optima. Evaluating the method on two benchmark models under various conditions, including absent low-frequency data, noise interference, and differing initial models, along with corresponding ablation experiments, consistently demonstrates the superiority of the proposed approach.
2502.11915
On the robustness of ChatGPT in teaching Korean Mathematics
cs.AI math.HO
ChatGPT, an Artificial Intelligence model, has the potential to revolutionize education. However, its effectiveness in solving non-English questions remains uncertain. This study evaluates ChatGPT's robustness using 586 Korean mathematics questions. ChatGPT achieves 66.72% accuracy, correctly answering 391 out of 586 questions. We also assess its ability to rate mathematics questions based on eleven criteria and perform a topic analysis. Our findings show that ChatGPT's ratings align with educational theory and test-taker perspectives. While ChatGPT performs well in question classification, it struggles with non-English contexts, highlighting areas for improvement. Future research should address linguistic biases and enhance accuracy across diverse languages. Domain-specific optimizations and multilingual training could improve ChatGPT's role in personalized education.
2502.11916
EssayJudge: A Multi-Granular Benchmark for Assessing Automated Essay Scoring Capabilities of Multimodal Large Language Models
cs.CL cs.AI
Automated Essay Scoring (AES) plays a crucial role in educational assessment by providing scalable and consistent evaluations of writing tasks. However, traditional AES systems face three major challenges: (1) reliance on handcrafted features that limit generalizability, (2) difficulty in capturing fine-grained traits like coherence and argumentation, and (3) inability to handle multimodal contexts. In the era of Multimodal Large Language Models (MLLMs), we propose EssayJudge, the first multimodal benchmark to evaluate AES capabilities across lexical-, sentence-, and discourse-level traits. By leveraging MLLMs' strengths in trait-specific scoring and multimodal context understanding, EssayJudge aims to offer precise, context-rich evaluations without manual feature engineering, addressing longstanding AES limitations. Our experiments with 18 representative MLLMs reveal gaps in AES performance compared to human evaluation, particularly in discourse-level traits, highlighting the need for further advancements in MLLM-based AES research. Our dataset and code will be available upon acceptance.
2502.11918
VLP: Vision-Language Preference Learning for Embodied Manipulation
cs.LG cs.RO
Reward engineering is one of the key challenges in Reinforcement Learning (RL). Preference-based RL effectively addresses this issue by learning from human feedback. However, it is both time-consuming and expensive to collect human preference labels. In this paper, we propose a novel \textbf{V}ision-\textbf{L}anguage \textbf{P}reference learning framework, named \textbf{VLP}, which learns a vision-language preference model to provide preference feedback for embodied manipulation tasks. To achieve this, we define three types of language-conditioned preferences and construct a vision-language preference dataset, which contains versatile implicit preference orders without human annotations. The preference model learns to extract language-related features, and then serves as a preference annotator in various downstream tasks. The policy can be learned according to the annotated preferences via reward learning or direct policy optimization. Extensive empirical results on simulated embodied manipulation tasks demonstrate that our method provides accurate preferences and generalizes to unseen tasks and unseen language instructions, outperforming the baselines by a large margin.
2502.11919
From Text to Trust: Empowering AI-assisted Decision Making with Adaptive LLM-powered Analysis
cs.HC cs.CL
AI-assisted decision making becomes increasingly prevalent, yet individuals often fail to utilize AI-based decision aids appropriately especially when the AI explanations are absent, potentially as they do not %understand reflect on AI's decision recommendations critically. Large language models (LLMs), with their exceptional conversational and analytical capabilities, present great opportunities to enhance AI-assisted decision making in the absence of AI explanations by providing natural-language-based analysis of AI's decision recommendation, e.g., how each feature of a decision making task might contribute to the AI recommendation. In this paper, via a randomized experiment, we first show that presenting LLM-powered analysis of each task feature, either sequentially or concurrently, does not significantly improve people's AI-assisted decision performance. To enable decision makers to better leverage LLM-powered analysis, we then propose an algorithmic framework to characterize the effects of LLM-powered analysis on human decisions and dynamically decide which analysis to present. Our evaluation with human subjects shows that this approach effectively improves decision makers' appropriate reliance on AI in AI-assisted decision making.
2502.11921
Joint Evaluation of Fairness and Relevance in Recommender Systems with Pareto Frontier
cs.IR
Fairness and relevance are two important aspects of recommender systems (RSs). Typically, they are evaluated either (i) separately by individual measures of fairness and relevance, or (ii) jointly using a single measure that accounts for fairness with respect to relevance. However, approach (i) often does not provide a reliable joint estimate of the goodness of the models, as it has two different best models: one for fairness and another for relevance. Approach (ii) is also problematic because these measures tend to be ad-hoc and do not relate well to traditional relevance measures, like NDCG. Motivated by this, we present a new approach for jointly evaluating fairness and relevance in RSs: Distance to Pareto Frontier (DPFR). Given some user-item interaction data, we compute their Pareto frontier for a pair of existing relevance and fairness measures, and then use the distance from the frontier as a measure of the jointly achievable fairness and relevance. Our approach is modular and intuitive as it can be computed with existing measures. Experiments with 4 RS models, 3 re-ranking strategies, and 6 datasets show that existing metrics have inconsistent associations with our Pareto-optimal solution, making DPFR a more robust and theoretically well-founded joint measure for assessing fairness and relevance. Our code: https://github.com/theresiavr/DPFR-recsys-evaluation
2502.11925
GRAPHGPT-O: Synergistic Multimodal Comprehension and Generation on Graphs
cs.AI cs.CV cs.LG
The rapid development of Multimodal Large Language Models (MLLMs) has enabled the integration of multiple modalities, including texts and images, within the large language model (LLM) framework. However, texts and images are usually interconnected, forming a multimodal attributed graph (MMAG). It is underexplored how MLLMs can incorporate the relational information (\textit{i.e.}, graph structure) and semantic information (\textit{i.e.,} texts and images) on such graphs for multimodal comprehension and generation. In this paper, we propose GraphGPT-o, which supports omni-multimodal understanding and creation on MMAGs. We first comprehensively study linearization variants to transform semantic and structural information as input for MLLMs. Then, we propose a hierarchical aligner that enables deep graph encoding, bridging the gap between MMAGs and MLLMs. Finally, we explore the inference choices, adapting MLLM to interleaved text and image generation in graph scenarios. Extensive experiments on three datasets from different domains demonstrate the effectiveness of our proposed method. Datasets and codes will be open-sourced upon acceptance.
2502.11926
BRIGHTER: BRIdging the Gap in Human-Annotated Textual Emotion Recognition Datasets for 28 Languages
cs.CL
People worldwide use language in subtle and complex ways to express emotions. While emotion recognition -- an umbrella term for several NLP tasks -- significantly impacts different applications in NLP and other fields, most work in the area is focused on high-resource languages. Therefore, this has led to major disparities in research and proposed solutions, especially for low-resource languages that suffer from the lack of high-quality datasets. In this paper, we present BRIGHTER -- a collection of multilabeled emotion-annotated datasets in 28 different languages. BRIGHTER covers predominantly low-resource languages from Africa, Asia, Eastern Europe, and Latin America, with instances from various domains annotated by fluent speakers. We describe the data collection and annotation processes and the challenges of building these datasets. Then, we report different experimental results for monolingual and crosslingual multi-label emotion identification, as well as intensity-level emotion recognition. We investigate results with and without using LLMs and analyse the large variability in performance across languages and text domains. We show that BRIGHTER datasets are a step towards bridging the gap in text-based emotion recognition and discuss their impact and utility.
2502.11927
Continual Learning Should Move Beyond Incremental Classification
cs.LG
Continual learning (CL) is the sub-field of machine learning concerned with accumulating knowledge in dynamic environments. So far, CL research has mainly focused on incremental classification tasks, where models learn to classify new categories while retaining knowledge of previously learned ones. Here, we argue that maintaining such a focus limits both theoretical development and practical applicability of CL methods. Through a detailed analysis of concrete examples - including multi-target classification, robotics with constrained output spaces, learning in continuous task domains, and higher-level concept memorization - we demonstrate how current CL approaches often fail when applied beyond standard classification. We identify three fundamental challenges: (C1) the nature of continuity in learning problems, (C2) the choice of appropriate spaces and metrics for measuring similarity, and (C3) the role of learning objectives beyond classification. For each challenge, we provide specific recommendations to help move the field forward, including formalizing temporal dynamics through distribution processes, developing principled approaches for continuous task spaces, and incorporating density estimation and generative objectives. In so doing, this position paper aims to broaden the scope of CL research while strengthening its theoretical foundations, making it more applicable to real-world problems.
2502.11932
On Representational Dissociation of Language and Arithmetic in Large Language Models
cs.CL
The association between language and (non-linguistic) thinking ability in humans has long been debated, and recently, neuroscientific evidence of brain activity patterns has been considered. Such a scientific context naturally raises an interdisciplinary question -- what about such a language-thought dissociation in large language models (LLMs)? In this paper, as an initial foray, we explore this question by focusing on simple arithmetic skills (e.g., $1+2=$ ?) as a thinking ability and analyzing the geometry of their encoding in LLMs' representation space. Our experiments with linear classifiers and cluster separability tests demonstrate that simple arithmetic equations and general language input are encoded in completely separated regions in LLMs' internal representation space across all the layers, which is also supported with more controlled stimuli (e.g., spelled-out equations). These tentatively suggest that arithmetic reasoning is mapped into a distinct region from general language input, which is in line with the neuroscientific observations of human brain activations, while we also point out their somewhat cognitively implausible geometric properties.
2502.11937
FitLight: Federated Imitation Learning for Plug-and-Play Autonomous Traffic Signal Control
cs.LG cs.AI
Although Reinforcement Learning (RL)-based Traffic Signal Control (TSC) methods have been extensively studied, their practical applications still raise some serious issues such as high learning cost and poor generalizability. This is because the ``trial-and-error'' training style makes RL agents extremely dependent on the specific traffic environment, which also requires a long convergence time. To address these issues, we propose a novel Federated Imitation Learning (FIL)-based framework for multi-intersection TSC, named FitLight, which allows RL agents to plug-and-play for any traffic environment without additional pre-training cost. Unlike existing imitation learning approaches that rely on pre-training RL agents with demonstrations, FitLight allows real-time imitation learning and seamless transition to reinforcement learning. Due to our proposed knowledge-sharing mechanism and novel hybrid pressure-based agent design, RL agents can quickly find a best control policy with only a few episodes. Moreover, for resource-constrained TSC scenarios, FitLight supports model pruning and heterogeneous model aggregation, such that RL agents can work on a micro-controller with merely 16{\it KB} RAM and 32{\it KB} ROM. Extensive experiments demonstrate that, compared to state-of-the-art methods, FitLight not only provides a superior starting point but also converges to a better final solution on both real-world and synthetic datasets, even under extreme resource limitations.
2502.11938
QoS based resource management for concurrent operation using MCTS
eess.SP cs.SY eess.SY
Modern AESA technology enables RF systems to not only perform various radar, communication and electronic warfare tasks on a single aperture, but even to execute multiple tasks concurrently. These capabilities increase system complexity and require intelligent or cognitive resource management. This paper introduces such a resource management framework based on quality of service based resource allocation and Monte Carlo tree search allowing for optimal system usage and profound decision-making. Furthermore, we present experimental verification in a complex application scenario.
2502.11940
The Dynamic Model of the UR10 Robot and its ROS2 Integration
cs.RO
This paper presents the full dynamic model of the UR10 industrial robot. A triple-stage identification approach is adopted to estimate the manipulator's dynamic coefficients. First, linear parameters are computed using a standard linear regression algorithm. Subsequently, nonlinear friction parameters are estimated according to a sigmoidal model. Lastly, motor drive gains are devised to map estimated joint currents to torques. The overall identified model can be used for both control and planning purposes, as the accompanied ROS2 software can be easily reconfigured to account for a generic payload. The estimated robot model is experimentally validated against a set of exciting trajectories and compared to the state-of-the-art model for the same manipulator, achieving higher current prediction accuracy (up to a factor of 4.43) and more precise motor gains. The related software is available at https://codeocean.com/capsule/8515919/tree/v2.
2502.11941
Deep Spatio-Temporal Neural Network for Air Quality Reanalysis
cs.LG cs.AI
Air quality prediction is key to mitigating health impacts and guiding decisions, yet existing models tend to focus on temporal trends while overlooking spatial generalization. We propose AQ-Net, a spatiotemporal reanalysis model for both observed and unobserved stations in the near future. AQ-Net utilizes the LSTM and multi-head attention for the temporal regression. We also propose a cyclic encoding technique to ensure continuous time representation. To learn fine-grained spatial air quality estimation, we incorporate AQ-Net with the neural kNN to explore feature-based interpolation, such that we can fill the spatial gaps given coarse observation stations. To demonstrate the efficiency of our model for spatiotemporal reanalysis, we use data from 2013-2017 collected in northern China for PM2.5 analysis. Extensive experiments show that AQ-Net excels in air quality reanalysis, highlighting the potential of hybrid spatio-temporal models to better capture environmental dynamics, especially in urban areas where both spatial and temporal variability are critical.
2502.11942
Sharp-PINNs: staggered hard-constrained physics-informed neural networks for phase field modelling of corrosion
cs.LG physics.comp-ph
Physics-informed neural networks have shown significant potential in solving partial differential equations (PDEs) across diverse scientific fields. However, their performance often deteriorates when addressing PDEs with intricate and strongly coupled solutions. In this work, we present a novel Sharp-PINN framework to tackle complex phase field corrosion problems. Instead of minimizing all governing PDE residuals simultaneously, the Sharp-PINNs introduce a staggered training scheme that alternately minimizes the residuals of Allen-Cahn and Cahn-Hilliard equations, which govern the corrosion system. To further enhance its efficiency and accuracy, we design an advanced neural network architecture that integrates random Fourier features as coordinate embeddings, employs a modified multi-layer perceptron as the primary backbone, and enforces hard constraints in the output layer. This framework is benchmarked through simulations of corrosion problems with multiple pits, where the staggered training scheme and network architecture significantly improve both the efficiency and accuracy of PINNs. Moreover, in three-dimensional cases, our approach is 5-10 times faster than traditional finite element methods while maintaining competitive accuracy, demonstrating its potential for real-world engineering applications in corrosion prediction.
2502.11946
Step-Audio: Unified Understanding and Generation in Intelligent Speech Interaction
cs.CL cs.AI cs.HC cs.SD eess.AS
Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.
2502.11948
Can Your Uncertainty Scores Detect Hallucinated Entity?
cs.CL
To mitigate the impact of hallucination nature of LLMs, many studies propose detecting hallucinated generation through uncertainty estimation. However, these approaches predominantly operate at the sentence or paragraph level, failing to pinpoint specific spans or entities responsible for hallucinated content. This lack of granularity is especially problematic for long-form outputs that mix accurate and fabricated information. To address this limitation, we explore entity-level hallucination detection. We propose a new data set, HalluEntity, which annotates hallucination at the entity level. Based on the dataset, we comprehensively evaluate uncertainty-based hallucination detection approaches across 17 modern LLMs. Our experimental results show that uncertainty estimation approaches focusing on individual token probabilities tend to over-predict hallucinations, while context-aware methods show better but still suboptimal performance. Through an in-depth qualitative study, we identify relationships between hallucination tendencies and linguistic properties and highlight important directions for future research.
2502.11949
Massively Scaling Explicit Policy-conditioned Value Functions
cs.LG cs.AI
We introduce a scaling strategy for Explicit Policy-Conditioned Value Functions (EPVFs) that significantly improves performance on challenging continuous-control tasks. EPVFs learn a value function V({\theta}) that is explicitly conditioned on the policy parameters, enabling direct gradient-based updates to the parameters of any policy. However, EPVFs at scale struggle with unrestricted parameter growth and efficient exploration in the policy parameter space. To address these issues, we utilize massive parallelization with GPU-based simulators, big batch sizes, weight clipping and scaled peturbations. Our results show that EPVFs can be scaled to solve complex tasks, such as a custom Ant environment, and can compete with state-of-the-art Deep Reinforcement Learning (DRL) baselines like Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC). We further explore action-based policy parameter representations from previous work and specialized neural network architectures to efficiently handle weight-space features, which have not been used in the context of DRL before.
2502.11951
Qubit-Based Framework for Quantum Machine Learning: Bridging Classical Data and Quantum Algorithms
cs.CE cs.LG quant-ph
This paper dives into the exciting and rapidly growing field of quantum computing, explaining its core ideas, current progress, and how it could revolutionize the way we solve complex problems. It starts by breaking down the basics, like qubits, quantum circuits, and how principles like superposition and entanglement make quantum computers fundamentally different-and far more powerful for certain tasks-than the classical computers we use today. We also explore how quantum computing deals with complex problems and why it is uniquely suited for challenges classical systems struggle to handle. A big part of this paper focuses on Quantum Machine Learning (QML), where the strengths of quantum computing meet the world of artificial intelligence. By processing massive datasets and optimizing intricate algorithms, quantum systems offer new possibilities for machine learning. We highlight different approaches to combining quantum and classical computing, showing how they can work together to produce faster and more accurate results. Additionally, we explore the tools and platforms available-like TensorFlow Quantum, Qiskit and PennyLane-that are helping researchers and developers bring these theories to life. Of course, quantum computing has its hurdles. Challenges like scaling up hardware, correcting errors, and keeping qubits stable are significant roadblocks. Yet, with rapid advancements in cloud-based platforms and innovative technologies, the potential of quantum computing feels closer than ever. This paper aims to offer readers a clear and comprehensive introduction to quantum computing, its role in machine learning, and the immense possibilities it holds for the future of technology.
2502.11953
Refined PAC-Bayes Bounds for Offline Bandits
stat.ML cs.LG
In this paper, we present refined probabilistic bounds on empirical reward estimates for off-policy learning in bandit problems. We build on the PAC-Bayesian bounds from Seldin et al. (2010) and improve on their results using a new parameter optimization approach introduced by Rodr\'iguez et al. (2024). This technique is based on a discretization of the space of possible events to optimize the "in probability" parameter. We provide two parameter-free PAC-Bayes bounds, one based on Hoeffding-Azuma's inequality and the other based on Bernstein's inequality. We prove that our bounds are almost optimal as they recover the same rate as would be obtained by setting the "in probability" parameter after the realization of the data.
2502.11955
pySLAM: An Open-Source, Modular, and Extensible Framework for SLAM
cs.RO cs.CV
pySLAM is an open-source Python framework for Visual SLAM, supporting monocular, stereo, and RGB-D cameras. It provides a flexible interface for integrating both classical and modern local features, making it adaptable to various SLAM tasks. The framework includes different loop closure methods, a volumetric reconstruction pipeline, and support for depth prediction models. Additionally, it offers a suite of tools for visual odometry and SLAM applications. Designed for both beginners and experienced researchers, pySLAM encourages community contributions, fostering collaborative development in the field of Visual SLAM.
2502.11959
STRIVE: Structured Reasoning for Self-Improvement in Claim Verification
cs.AI
Claim verification is the task of determining whether a claim is supported or refuted by evidence. Self-improvement methods, where reasoning chains are generated and those leading to correct results are selected for training, have succeeded in tasks like mathematical problem solving. However, in claim verification, this approach struggles. Low-quality reasoning chains may falsely match binary truth labels, introducing faulty reasoning into the self-improvement process and ultimately degrading performance. To address this, we propose STRIVE: Structured Reasoning for Self-Improved Verification. Our method introduces a structured reasoning design with Claim Decomposition, Entity Analysis, and Evidence Grounding Verification. These components improve reasoning quality, reduce errors, and provide additional supervision signals for self-improvement. STRIVE begins with a warm-up phase, where the base model is fine-tuned on a small number of annotated examples to learn the structured reasoning design. It is then applied to generate reasoning chains for all training examples, selecting only those that are correct and structurally sound for subsequent self-improvement training. We demonstrate that STRIVE achieves significant improvements over baseline models, with a 31.4% performance gain over the base model and 20.7% over Chain of Thought on the HOVER datasets, highlighting its effectiveness.
2502.11962
Navigating the Helpfulness-Truthfulness Trade-Off with Uncertainty-Aware Instruction Fine-Tuning
cs.CL cs.AI
Instruction Fine-tuning (IFT) can enhance the helpfulness of Large Language Models (LLMs), but it may lower their truthfulness. This trade-off arises because IFT steers LLMs to generate responses with long-tail knowledge that is not well covered during pre-training, leading to more informative but less truthful answers when generalizing to unseen tasks. In this paper, we empirically demonstrate this helpfulness-truthfulness trade-off in IFT and propose $\textbf{UNIT}$, a novel IFT paradigm to address it. UNIT teaches LLMs to recognize their uncertainty and explicitly reflect it at the end of their responses. Experimental results show that UNIT-tuned models maintain their helpfulness while distinguishing between certain and uncertain claims, thereby reducing hallucinations.
2502.11965
A MIMO Wireless Channel Foundation Model via CIR-CSI Consistency
eess.SP cs.AI
In the field of artificial intelligence, self-supervised learning has demonstrated superior generalization capabilities by leveraging large-scale unlabeled datasets for pretraining, which is especially critical for wireless communication models to adapt to a variety of scenarios. This paper innovatively treats Channel State Information (CSI) and Channel Impulse Response (CIR) as naturally aligned multi-modal data and proposes the first MIMO wireless channel foundation model, named CSI-CLIP. By effectively capturing the joint representations of both CIR and CSI, CSI-CLIP exhibits remarkable adaptability across scenarios and robust feature extraction capabilities. Experimental results show that in positioning task, CSI-CLIP reduces the mean error distance by 22%; in beam management task, it increases accuracy by 1% compared to traditional supervised methods, as well as in the channel identification task. These improvements not only highlight the potential and value of CSI-CLIP in integrating sensing and communication but also demonstrate its significant advantages over existing techniques. Moreover, viewing CSI and CIR as multi-modal pairs and contrastive learning for wireless channel foundation model open up new research directions in the domain of MIMO wireless communications.
2502.11968
Theoretical Barriers in Bellman-Based Reinforcement Learning
cs.LG cs.AI
Reinforcement Learning algorithms designed for high-dimensional spaces often enforce the Bellman equation on a sampled subset of states, relying on generalization to propagate knowledge across the state space. In this paper, we identify and formalize a fundamental limitation of this common approach. Specifically, we construct counterexample problems with a simple structure that this approach fails to exploit. Our findings reveal that such algorithms can neglect critical information about the problems, leading to inefficiencies. Furthermore, we extend this negative result to another approach from the literature: Hindsight Experience Replay learning state-to-state reachability.
2502.11969
Learning Generalizable Prompt for CLIP with Class Similarity Knowledge
cs.AI cs.CV cs.LG
In vision-language models (VLMs), prompt tuning has shown its effectiveness in adapting models to downstream tasks. However, learned prompts struggle to generalize to unseen classes, as they tend to overfit to the classes that are targeted during prompt tuning. Examining failure cases, we observed that learned prompts disrupt the semantics of unseen classes, generating text embeddings with incorrect semantic relationships among classes. To address this, we propose Similarity Alignment Regularization (SAR), which regularizes learnable prompts to preserve the semantic relationships among classes captured by hand-crafted prompts. Specifically, we first obtain novel classes related to base classes using ChatGPT-4o and utilize them as potential unseen classes during prompt tuning. Then, by targeting both base and novel classes, SAR aligns the similarity relationships among text embeddings generated by learnable prompts with the similarity relationships from hand-crafted prompts. Extensive experiments applying SAR to existing prompt tuning methods demonstrate its effectiveness in improving generalization to unseen classes.
2502.11971
Robust 6DoF Pose Tracking Considering Contour and Interior Correspondence Uncertainty for AR Assembly Guidance
cs.CV
Augmented reality assembly guidance is essential for intelligent manufacturing and medical applications, requiring continuous measurement of the 6DoF poses of manipulated objects. Although current tracking methods have made significant advancements in accuracy and efficiency, they still face challenges in robustness when dealing with cluttered backgrounds, rotationally symmetric objects, and noisy sequences. In this paper, we first propose a robust contour-based pose tracking method that addresses error-prone contour correspondences and improves noise tolerance. It utilizes a fan-shaped search strategy to refine correspondences and models local contour shape and noise uncertainty as mixed probability distribution, resulting in a highly robust contour energy function. Secondly, we introduce a CPU-only strategy to better track rotationally symmetric objects and assist the contour-based method in overcoming local minima by exploring sparse interior correspondences. This is achieved by pre-sampling interior points from sparse viewpoint templates offline and using the DIS optical flow algorithm to compute their correspondences during tracking. Finally, we formulate a unified energy function to fuse contour and interior information, which is solvable using a re-weighted least squares algorithm. Experiments on public datasets and real scenarios demonstrate that our method significantly outperforms state-of-the-art monocular tracking methods and can achieve more than 100 FPS using only a CPU.
2502.11973
Generating Text from Uniform Meaning Representation
cs.CL
Uniform Meaning Representation (UMR) is a recently developed graph-based semantic representation, which expands on Abstract Meaning Representation (AMR) in a number of ways, in particular through the inclusion of document-level information and multilingual flexibility. In order to effectively adopt and leverage UMR for downstream tasks, efforts must be placed toward developing a UMR technological ecosystem. Though still limited amounts of UMR annotations have been produced to date, in this work, we investigate the first approaches to producing text from multilingual UMR graphs: (1) a pipeline conversion of UMR to AMR, then using AMR-to-text generation models, (2) fine-tuning large language models with UMR data, and (3) fine-tuning existing AMR-to-text generation models with UMR data. Our best performing model achieves a multilingual BERTscore of 0.825 for English and 0.882 for Chinese when compared to the reference, which is a promising indication of the effectiveness of fine-tuning approaches for UMR-to-text generation with even limited amounts of UMR data.
2502.11974
Image Inversion: A Survey from GANs to Diffusion and Beyond
cs.CV
Image inversion is a fundamental task in generative models, aiming to map images back to their latent representations to enable downstream applications such as editing, restoration, and style transfer. This paper provides a comprehensive review of the latest advancements in image inversion techniques, focusing on two main paradigms: Generative Adversarial Network (GAN) inversion and diffusion model inversion. We categorize these techniques based on their optimization methods. For GAN inversion, we systematically classify existing methods into encoder-based approaches, latent optimization approaches, and hybrid approaches, analyzing their theoretical foundations, technical innovations, and practical trade-offs. For diffusion model inversion, we explore training-free strategies, fine-tuning methods, and the design of additional trainable modules, highlighting their unique advantages and limitations. Additionally, we discuss several popular downstream applications and emerging applications beyond image tasks, identifying current challenges and future research directions. By synthesizing the latest developments, this paper aims to provide researchers and practitioners with a valuable reference resource, promoting further advancements in the field of image inversion. We keep track of the latest works at https://github.com/RyanChenYN/ImageInversion
2502.11975
Spatial decay of perturbations in hyperbolic equations with optimal boundary control
math.OC cs.SY eess.SY
Recently, domain-uniform stabilizability and detectability has been the central assumption %in order robustness results on the to ensure robustness in the sense of exponential decay of spatially localized perturbations in optimally controlled evolution equations. In the present paper we analyze a chain of transport equations with boundary and point controls with regard to this property. Both for Dirichlet and Neumann boundary and coupling conditions, we show a necessary and sufficient criterion on control domains which allow for the domain-uniform stabilization of this equation. We illustrate the results by means of a numerical example.
2502.11981
Machine Learning Should Maximize Welfare, Not (Only) Accuracy
cs.LG cs.AI cs.CY
Decades of research in machine learning have given us powerful tools for making accurate predictions. But when used in social settings and on human inputs, better accuracy does not immediately translate to better social outcomes. This may not be surprising given that conventional learning frameworks are not designed to express societal preferences -- let alone promote them. This position paper argues that machine learning is currently missing, and can gain much from incorporating, a proper notion of social welfare. The field of welfare economics asks: how should we allocate limited resources to self-interested agents in a way that maximizes social benefit? We argue that this perspective applies to many modern applications of machine learning in social contexts, and advocate for its adoption. Rather than disposing of prediction, we aim to leverage this forte of machine learning for promoting social welfare. We demonstrate this idea by proposing a conceptual framework that gradually transitions from accuracy maximization (with awareness to welfare) to welfare maximization (via accurate prediction). We detail applications and use-cases for which our framework can be effective, identify technical challenges and practical opportunities, and highlight future avenues worth pursuing.
2502.11983
Design Considerations Based on Stability for a Class of TCP Algorithms
cs.NI cs.SY eess.SY
Transmission Control Protocol (TCP) continues to be the dominant transport protocol on the Internet. The stability of fluid models has been a key consideration in the design of TCP and the performance evaluation of TCP algorithms. Based on local stability analysis, we formulate some design considerations for a class of TCP algorithms. We begin with deriving sufficient conditions for the local stability of a generalized TCP algorithm in the presence of heterogeneous round-trip delays. Within this generalized model, we consider three specific variants of TCP: TCP Reno, Compound TCP, and Scalable TCP. The sufficient conditions we derive are scalable across network topologies with one, two, and many bottleneck links. We are interested in networks with intermediate and small drop-tail buffers as they offer smaller queuing delays. The small buffer regime is more attractive as the conditions for stability are decentralized. TCP algorithms that follow our design considerations can provide stable operation on any network topology, irrespective of the number of bottleneck links or delays in the network.
2502.11984
Blank Space: Adaptive Causal Coding for Streaming Communications Over Multi-Hop Networks
cs.IT cs.NI math.IT
In this work, we introduce Blank Space AC-RLNC (BS), a novel Adaptive and Causal Network Coding (AC-RLNC) solution designed to mitigate the triplet trade-off between throughput-delay-efficiency in multi-hop networks. BS leverages the network's physical limitations considering the bottleneck from each node to the destination. In particular, BS introduces a light-computational re-encoding algorithm, called Network AC-RLNC (NET), implemented independently at intermediate nodes. NET adaptively adjusts the Forward Error Correction (FEC) rates and schedules idle periods. It incorporates two distinct suspension mechanisms: 1) Blank Space Period, accounting for the forward-channels bottleneck, and 2) No-New No-FEC approach, based on data availability. The experimental results achieve significant improvements in resource efficiency, demonstrating a 20% reduction in channel usage compared to baseline RLNC solutions. Notably, these efficiency gains are achieved while maintaining competitive throughput and delay performance, ensuring improved resource utilization does not compromise network performance.
2502.11986
Selective Task Group Updates for Multi-Task Optimization
cs.LG
Multi-task learning enables the acquisition of task-generic knowledge by training multiple tasks within a unified architecture. However, training all tasks together in a single architecture can lead to performance degradation, known as negative transfer, which is a main concern in multi-task learning. Previous works have addressed this issue by optimizing the multi-task network through gradient manipulation or weighted loss adjustments. However, their optimization strategy focuses on addressing task imbalance in shared parameters, neglecting the learning of task-specific parameters. As a result, they show limitations in mitigating negative transfer, since the learning of shared space and task-specific information influences each other during optimization. To address this, we propose a different approach to enhance multi-task performance by selectively grouping tasks and updating them for each batch during optimization. We introduce an algorithm that adaptively determines how to effectively group tasks and update them during the learning process. To track inter-task relations and optimize multi-task networks simultaneously, we propose proximal inter-task affinity, which can be measured during the optimization process. We provide a theoretical analysis on how dividing tasks into multiple groups and updating them sequentially significantly affects multi-task performance by enhancing the learning of task-specific parameters. Our methods substantially outperform previous multi-task optimization approaches and are scalable to different architectures and various numbers of tasks.
2502.11989
Characterizing Photorealism and Artifacts in Diffusion Model-Generated Images
cs.HC cs.AI cs.CV
Diffusion model-generated images can appear indistinguishable from authentic photographs, but these images often contain artifacts and implausibilities that reveal their AI-generated provenance. Given the challenge to public trust in media posed by photorealistic AI-generated images, we conducted a large-scale experiment measuring human detection accuracy on 450 diffusion-model generated images and 149 real images. Based on collecting 749,828 observations and 34,675 comments from 50,444 participants, we find that scene complexity of an image, artifact types within an image, display time of an image, and human curation of AI-generated images all play significant roles in how accurately people distinguish real from AI-generated images. Additionally, we propose a taxonomy characterizing artifacts often appearing in images generated by diffusion models. Our empirical observations and taxonomy offer nuanced insights into the capabilities and limitations of diffusion models to generate photorealistic images in 2024.
2502.11992
On the Logic Elements Associated with Round-Off Errors and Gaussian Blur in Image Registration: A Simple Case of Commingling
cs.CV
Discrete image registration can be a strategy to reconstruct signals from samples corrupted by blur and noise. We examine superresolution and discrete image registration for one-dimensional spatially-limited piecewise constant functions which are subject to blur which is Gaussian or a mixture of Gaussians as well as to round-off errors. Previous approaches address the signal recovery problem as an optimization problem. We focus on a regime with low blur and suggest that the operations of blur, sampling, and quantization are not unlike the operation of a computer program and have an abstraction that can be studied with a type of logic. When the minimum distance between discontinuity points is between $1.5$ and 2 times the sampling interval, we can encounter the simplest form of a type of interference between discontinuity points that we call ``commingling.'' We describe a way to reason about two sets of samples of the same signal that will often result in the correct recovery of signal amplitudes. We also discuss ways to estimate bounds on the distances between discontinuity points.
2502.11993
MultiFlow: A unified deep learning framework for multi-vessel classification, segmentation and clustering of phase-contrast MRI validated on a multi-site single ventricle patient cohort
cs.CV
This study presents a unified deep learning (DL) framework, MultiFlowSeg, for classification and segmentation of velocity-encoded phase-contrast magnetic resonance imaging data, and MultiFlowDTC for temporal clustering of flow phenotypes. Applied to the FORCE registry of Fontan procedure patients, MultiFlowSeg achieved 100% classification accuracy for the aorta, SVC, and IVC, and 94% for the LPA and RPA. It demonstrated robust segmentation with a median Dice score of 0.91 (IQR: 0.86-0.93). The automated pipeline processed registry data, achieving high segmentation success despite challenges like poor image quality and dextrocardia. Temporal clustering identified five distinct patient subgroups, with significant differences in clinical outcomes, including ejection fraction, exercise tolerance, liver disease, and mortality. These results demonstrate the potential of combining DL and time-varying flow data for improved CHD prognosis and personalized care.
2502.11995
Presumed Cultural Identity: How Names Shape LLM Responses
cs.CL cs.AI
Names are deeply tied to human identity. They can serve as markers of individuality, cultural heritage, and personal history. However, using names as a core indicator of identity can lead to over-simplification of complex identities. When interacting with LLMs, user names are an important point of information for personalisation. Names can enter chatbot conversations through direct user input (requested by chatbots), as part of task contexts such as CV reviews, or as built-in memory features that store user information for personalisation. We study biases associated with names by measuring cultural presumptions in the responses generated by LLMs when presented with common suggestion-seeking queries, which might involve making assumptions about the user. Our analyses demonstrate strong assumptions about cultural identity associated with names present in LLM generations across multiple cultures. Our work has implications for designing more nuanced personalisation systems that avoid reinforcing stereotypes while maintaining meaningful customisation.
2502.12001
Merging Language and Domain Specific Models: The Impact on Technical Vocabulary Acquisition
cs.CL cs.LG
This paper investigates the integration of technical vocabulary in merged language models. We explore the knowledge transfer mechanisms involved when combining a general-purpose language-specific model with a domain-specific model, focusing on the resulting model's comprehension of technical jargon. Our experiments analyze the impact of this merging process on the target model's proficiency in handling specialized terminology. We present a quantitative evaluation of the performance of the merged model, comparing it with that of the individual constituent models. The findings offer insights into the effectiveness of different model merging methods for enhancing domain-specific knowledge and highlight potential challenges and future directions in leveraging these methods for cross-lingual knowledge transfer in Natural Language Processing.
2502.12002
NaturalL2S: End-to-End High-quality Multispeaker Lip-to-Speech Synthesis with Differential Digital Signal Processing
cs.SD cs.CV eess.AS
Recent advancements in visual speech recognition (VSR) have promoted progress in lip-to-speech synthesis, where pre-trained VSR models enhance the intelligibility of synthesized speech by providing valuable semantic information. The success achieved by cascade frameworks, which combine pseudo-VSR with pseudo-text-to-speech (TTS) or implicitly utilize the transcribed text, highlights the benefits of leveraging VSR models. However, these methods typically rely on mel-spectrograms as an intermediate representation, which may introduce a key bottleneck: the domain gap between synthetic mel-spectrograms, generated from inherently error-prone lip-to-speech mappings, and real mel-spectrograms used to train vocoders. This mismatch inevitably degrades synthesis quality. To bridge this gap, we propose Natural Lip-to-Speech (NaturalL2S), an end-to-end framework integrating acoustic inductive biases with differentiable speech generation components. Specifically, we introduce a fundamental frequency (F0) predictor to capture prosodic variations in synthesized speech. The predicted F0 then drives a Differentiable Digital Signal Processing (DDSP) synthesizer to generate a coarse signal which serves as prior information for subsequent speech synthesis. Additionally, instead of relying on a reference speaker embedding as an auxiliary input, our approach achieves satisfactory performance on speaker similarity without explicitly modelling speaker characteristics. Both objective and subjective evaluation results demonstrate that NaturalL2S can effectively enhance the quality of the synthesized speech when compared to state-of-the-art methods. Our demonstration page is accessible at https://yifan-liang.github.io/NaturalL2S/.
2502.12003
Predicting Next-Day Wildfire Spread with Time Series and Attention
cs.CV
Recent research has demonstrated the potential of deep neural networks (DNNs) to accurately predict next-day wildfire spread, based upon the current extent of a fire and geospatial rasters of influential environmental covariates e.g., vegetation, topography, climate, and weather. In this work, we investigate a recent transformer-based model, termed the SwinUnet, for next-day wildfire prediction. We benchmark Swin-based models against several current state-of-the-art models on WildfireSpreadTS (WFTS), a large public benchmark dataset of historical wildfire events. We consider two next-day fire prediction scenarios: when the model is given input of (i) a single previous day of data, or (ii) five previous days of data. We find that, with the proper modifications, SwinUnet achieves state-of-the-art accuracy on next-day prediction for both the single-day and multi-day scenarios. SwinUnet's success depends heavily upon utilizing pre-trained weights from ImageNet. Consistent with prior work, we also found that models with multi-day-input always outperformed models with single-day input.
2502.12005
Feasibility Evaluation of Quadratic Programs for Constrained Control
math.OC cs.SY eess.SY
This paper presents a computationally-efficient method for evaluating the feasibility of Quadratic Programs (QPs) for online constrained control. Based on the duality principle, we first show that the feasibility of a QP can be determined by the solution of a properly-defined Linear Program (LP). Our analysis yields a LP that can be solved more efficiently compared to the original QP problem, and more importantly, is simpler in form and can be solved more efficiently compared to existing methods that assess feasibility via LPs. The computational efficiency of the proposed method compared to existing methods for feasibility evaluation is demonstrated in comparative case studies as well as a feasible-constraint selection problem, indicating its promise for online feasibility evaluation of optimization-based controllers.
2502.12007
Demographic Attributes Prediction from Speech Using WavLM Embeddings
cs.CL cs.AI
This paper introduces a general classifier based on WavLM features, to infer demographic characteristics, such as age, gender, native language, education, and country, from speech. Demographic feature prediction plays a crucial role in applications like language learning, accessibility, and digital forensics, enabling more personalized and inclusive technologies. Leveraging pretrained models for embedding extraction, the proposed framework identifies key acoustic and linguistic fea-tures associated with demographic attributes, achieving a Mean Absolute Error (MAE) of 4.94 for age prediction and over 99.81% accuracy for gender classification across various datasets. Our system improves upon existing models by up to relative 30% in MAE and up to relative 10% in accuracy and F1 scores across tasks, leveraging a diverse range of datasets and large pretrained models to ensure robustness and generalizability. This study offers new insights into speaker diversity and provides a strong foundation for future research in speech-based demographic profiling.
2502.12009
Beyond Sentiment: Examining the Role of Moral Foundations in User Engagement with News on Twitter
cs.SI
This study uses sentiment analysis and the Moral Foundations Theory (MFT) to characterise news content in social media and examine its association with user engagement. We employ Natural Language Processing to quantify the moral and affective linguistic markers. At the same time, we automatically define thematic macro areas of news from major U.S. news outlets and their Twitter followers (Jan 2020 - Mar 2021). By applying Non-Negative Matrix Factorisation to the obtained linguistic features we extract clusters of similar moral and affective profiles, and we identify the emotional and moral characteristics that mostly explain user engagement via regression modelling. We observe that Surprise, Trust, and Harm are crucial elements explaining user engagement and discussion length and that Twitter content from news media outlets has more explanatory power than their linked articles. We contribute with actionable findings evidencing the potential impact of employing specific moral and affective nuances in public and journalistic discourse in today's communication landscape. In particular, our results emphasise the need to balance engagement strategies with potential priming risks in our evolving media landscape.
2502.12011
Reconfigurable Intelligent Surfaces-Assisted Integrated Access and Backhaul
cs.IT cs.LG cs.NI math.IT
In this paper, we study the impact of reconfigurable intelligent surfaces (RISs) on the coverage extension of integrated access and backhaul (IAB) networks. Particularly, using a finite stochastic geometry model, with random distributions of user equipments (UEs) in a finite region, and planned hierachical architecture for IAB, we study the service coverage probability defined as the probability of the event that the UEs' minimum rate requirements are satisfied. We present comparisons between different cases including IAB-only, IAB assisted with RIS for backhaul as well as IAB assisted by network controlled repeaters (NCRs). Our investigations focus on wide-area IAB assisted with RIS through the lens of different design architectures and deployments, revealing both conflicts and synergies for minimizing the effect of tree foliage over seasonal changes. Our simulation results reveal both opportunities and challenges towards the implementation of RIS in IAB.
2502.12012
Evolving Hard Maximum Cut Instances for Quantum Approximate Optimization Algorithms
cs.ET cs.AI cs.NE quant-ph
Variational quantum algorithms, such as the Recursive Quantum Approximate Optimization Algorithm (RQAOA), have become increasingly popular, offering promising avenues for employing Noisy Intermediate-Scale Quantum devices to address challenging combinatorial optimization tasks like the maximum cut problem. In this study, we utilize an evolutionary algorithm equipped with a unique fitness function. This approach targets hard maximum cut instances within the latent space of a Graph Autoencoder, identifying those that pose significant challenges or are particularly tractable for RQAOA, in contrast to the classic Goemans and Williamson algorithm. Our findings not only delineate the distinct capabilities and limitations of each algorithm but also expand our understanding of RQAOA's operational limits. Furthermore, the diverse set of graphs we have generated serves as a crucial benchmarking asset, emphasizing the need for more advanced algorithms to tackle combinatorial optimization challenges. Additionally, our results pave the way for new avenues in graph generation research, offering exciting opportunities for future explorations.
2502.12013
Unsupervised Structural-Counterfactual Generation under Domain Shift
cs.LG stat.ML
Motivated by the burgeoning interest in cross-domain learning, we present a novel generative modeling challenge: generating counterfactual samples in a target domain based on factual observations from a source domain. Our approach operates within an unsupervised paradigm devoid of parallel or joint datasets, relying exclusively on distinct observational samples and causal graphs for each domain. This setting presents challenges that surpass those of conventional counterfactual generation. Central to our methodology is the disambiguation of exogenous causes into effect-intrinsic and domain-intrinsic categories. This differentiation facilitates the integration of domain-specific causal graphs into a unified joint causal graph via shared effect-intrinsic exogenous variables. We propose leveraging Neural Causal models within this joint framework to enable accurate counterfactual generation under standard identifiability assumptions. Furthermore, we introduce a novel loss function that effectively segregates effect-intrinsic from domain-intrinsic variables during model training. Given a factual observation, our framework combines the posterior distribution of effect-intrinsic variables from the source domain with the prior distribution of domain-intrinsic variables from the target domain to synthesize the desired counterfactuals, adhering to Pearl's causal hierarchy. Intriguingly, when domain shifts are restricted to alterations in causal mechanisms without accompanying covariate shifts, our training regimen parallels the resolution of a conditional optimal transport problem. Empirical evaluations on a synthetic dataset show that our framework generates counterfactuals in the target domain that very closely resemble the ground truth.
2502.12017
Scalable and Cost-Efficient ML Inference: Parallel Batch Processing with Serverless Functions
cs.DC cs.LG
As data-intensive applications grow, batch processing in limited-resource environments faces scalability and resource management challenges. Serverless computing offers a flexible alternative, enabling dynamic resource allocation and automatic scaling. This paper explores how serverless architectures can make large-scale ML inference tasks faster and cost-effective by decomposing monolithic processes into parallel functions. Through a case study on sentiment analysis using the DistilBERT model and the IMDb dataset, we demonstrate that serverless parallel processing can reduce execution time by over 95% compared to monolithic approaches, at the same cost.
2502.12018
Atom of Thoughts for Markov LLM Test-Time Scaling
cs.CL cs.AI cs.LG
Large Language Models (LLMs) achieve superior performance through training-time scaling, and test-time scaling further enhances their capabilities by conducting effective reasoning during inference. However, as the scale of reasoning increases, existing test-time scaling methods suffer from accumulated historical information, which not only wastes computational resources but also interferes with effective reasoning. To address this issue, we observe that complex reasoning progress is often achieved by solving a sequence of independent subquestions, each being self-contained and verifiable. These subquestions are essentially atomic questions, relying primarily on their current state rather than accumulated history, similar to the memoryless transitions in a Markov process. Based on this observation, we propose Atom of Thoughts (AoT), where each state transition in the reasoning process consists of decomposing the current question into a dependency-based directed acyclic graph and contracting its subquestions, forming a new atomic question state. This iterative decomposition-contraction process continues until reaching directly solvable atomic questions, naturally realizing Markov transitions between question states. Furthermore, these atomic questions can be seamlessly integrated into existing test-time scaling methods, enabling AoT to serve as a plug-in enhancement for improving reasoning capabilities. Experiments across six benchmarks demonstrate the effectiveness of AoT both as a standalone framework and a plug-in enhancement. Notably, on HotpotQA, when applied to gpt-4o-mini, AoT achieves an 80.6% F1 score, surpassing o3-mini by 3.4% and DeepSeek-R1 by 10.6%. The code will be available at https://github.com/qixucen/atom.
2502.12019
Robotic CBCT Meets Robotic Ultrasound
cs.RO eess.IV
The multi-modality imaging system offers optimal fused images for safe and precise interventions in modern clinical practices, such as computed tomography - ultrasound (CT-US) guidance for needle insertion. However, the limited dexterity and mobility of current imaging devices hinder their integration into standardized workflows and the advancement toward fully autonomous intervention systems. In this paper, we present a novel clinical setup where robotic cone beam computed tomography (CBCT) and robotic US are pre-calibrated and dynamically co-registered, enabling new clinical applications. This setup allows registration-free rigid registration, facilitating multi-modal guided procedures in the absence of tissue deformation. First, a one-time pre-calibration is performed between the systems. To ensure a safe insertion path by highlighting critical vasculature on the 3D CBCT, SAM2 segments vessels from B-mode images, using the Doppler signal as an autonomously generated prompt. Based on the registration, the Doppler image or segmented vessel masks are then mapped onto the CBCT, creating an optimally fused image with comprehensive detail. To validate the system, we used a specially designed phantom, featuring lesions covered by ribs and multiple vessels with simulated moving flow. The mapping error between US and CBCT resulted in an average deviation of 1.72+-0.62 mm. A user study demonstrated the effectiveness of CBCT-US fusion for needle insertion guidance, showing significant improvements in time efficiency, accuracy, and success rate. Needle intervention performance improved by approximately 50% compared to the conventional US-guided workflow. We present the first robotic dual-modality imaging system designed to guide clinical applications. The results show significant performance improvements compared to traditional manual interventions.
2502.12020
Learning in a Multifield Coherent Ising Machine
cond-mat.mes-hall cond-mat.dis-nn cs.ET cs.NE nlin.AO
Physical information processors can learn from examples if they are modified according to an abstract parameter update equation, termed a learning rule. We introduce a physical model for self-learning that encodes the learning rule in the Hamiltonian of the system. The model consists of a network of multi-modal resonators. One of the modes is driven parametrically into a bi-stable regime, forming a coherent Ising machine (CIM) -- that provides the long-term memory that stores learned responses (weights). The CIM is augmented with an additional spinor field that acts as short-term (activation) memory. We numerically demonstrate that, in the presence of suitable nonlinear interactions between the long-term memory Ising machine and the short-term memory auxiliary field, the system autonomously learns from examples.
2502.12022
Teaching LLMs According to Their Aptitude: Adaptive Reasoning for Mathematical Problem Solving
cs.CL cs.AI
Existing approaches to mathematical reasoning with large language models (LLMs) rely on Chain-of-Thought (CoT) for generalizability or Tool-Integrated Reasoning (TIR) for precise computation. While efforts have been made to combine these methods, they primarily rely on post-selection or predefined strategies, leaving an open question: whether LLMs can autonomously adapt their reasoning strategy based on their inherent capabilities. In this work, we propose TATA (Teaching LLMs According to Their Aptitude), an adaptive framework that enables LLMs to personalize their reasoning strategy spontaneously, aligning it with their intrinsic aptitude. TATA incorporates base-LLM-aware data selection during supervised fine-tuning (SFT) to tailor training data to the model's unique abilities. This approach equips LLMs to autonomously determine and apply the appropriate reasoning strategy at test time. We evaluate TATA through extensive experiments on six mathematical reasoning benchmarks, using both general-purpose and math-specialized LLMs. Empirical results demonstrate that TATA effectively combines the complementary strengths of CoT and TIR, achieving superior or comparable performance with improved inference efficiency compared to TIR alone. Further analysis underscores the critical role of aptitude-aware data selection in enabling LLMs to make effective and adaptive reasoning decisions and align reasoning strategies with model capabilities.
2502.12025
SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities
cs.AI cs.CL
Emerging large reasoning models (LRMs), such as DeepSeek-R1 models, leverage long chain-of-thought (CoT) reasoning to generate structured intermediate steps, enhancing their reasoning capabilities. However, long CoT does not inherently guarantee safe outputs, potentially leading to harmful consequences such as the introduction of security vulnerabilities in code or the spread of misinformation. Current research on large language model (LLM) safety usually focuses on short-answer responses, overlooking the long CoT style outputs of LRMs. To bridge this gap, we conduct a systematic study of LRM safety. First, we investigate safety evaluators calibrated against human annotations. Using our newly developed metrics, we thoroughly assess the safety of 12 state-of-the-art LRMs on StrongReject and WildJailbreak datasets. Our results show that LRMs are not safe compared to their reasoning advance. Further, we perform a fine-grained analysis of the reasoning trace and final answer. We find that three decoding strategies-ZeroThink, LessThink, and MoreThink-can improve model safety without additional training. However, these strategies either use constrained reasoning traces or incur high inference costs. To better strengthen LRM safety, we introduce SafeChain, the first-of-its-kind safety training dataset in CoT style. We fine-tune two LRMs with SafeChain, showing that it not only enhances model safety but also preserves performance across 6 reasoning benchmarks.
2502.12027
Enhancing Transparent Object Pose Estimation: A Fusion of GDR-Net and Edge Detection
cs.CV
Object pose estimation of transparent objects remains a challenging task in the field of robot vision due to the immense influence of lighting, background, and reflections. However, the edges of clear objects have the highest contrast, which leads to stable and prominent features. We propose a novel approach by incorporating edge detection in a pre-processing step for the tasks of object detection and object pose estimation. We conducted experiments to investigate the effect of edge detectors on transparent objects. We examine the performance of the state-of-the-art 6D object pose estimation pipeline GDR-Net and the object detector YOLOX when applying different edge detectors as pre-processing steps (i.e., Canny edge detection with and without color information, and holistically-nested edges (HED)). We evaluate the physically-based rendered dataset Trans6D-32 K of transparent objects with parameters proposed by the BOP Challenge. Our results indicate that applying edge detection as a pre-processing enhances performance for certain objects.
2502.12029
KnowPath: Knowledge-enhanced Reasoning via LLM-generated Inference Paths over Knowledge Graphs
cs.AI
Large language models (LLMs) have demonstrated remarkable capabilities in various complex tasks, yet they still suffer from hallucinations. Introducing external knowledge, such as knowledge graph, can enhance the LLMs' ability to provide factual answers. LLMs have the ability to interactively explore knowledge graphs. However, most approaches have been affected by insufficient internal knowledge excavation in LLMs, limited generation of trustworthy knowledge reasoning paths, and a vague integration between internal and external knowledge. Therefore, we propose KnowPath, a knowledge-enhanced large model framework driven by the collaboration of internal and external knowledge. It relies on the internal knowledge of the LLM to guide the exploration of interpretable directed subgraphs in external knowledge graphs, better integrating the two knowledge sources for more accurate reasoning. Extensive experiments on multiple real-world datasets confirm the superiority of KnowPath.
2502.12031
Masked Latent Prediction and Classification for Self-Supervised Audio Representation Learning
cs.SD cs.AI
Recently, self-supervised learning methods based on masked latent prediction have proven to encode input data into powerful representations. However, during training, the learned latent space can be further transformed to extract higher-level information that could be more suited for downstream classification tasks. Therefore, we propose a new method: MAsked latenT Prediction And Classification (MATPAC), which is trained with two pretext tasks solved jointly. As in previous work, the first pretext task is a masked latent prediction task, ensuring a robust input representation in the latent space. The second one is unsupervised classification, which utilises the latent representations of the first pretext task to match probability distributions between a teacher and a student. We validate the MATPAC method by comparing it to other state-of-the-art proposals and conducting ablations studies. MATPAC reaches state-of-the-art self-supervised learning results on reference audio classification datasets such as OpenMIC, GTZAN, ESC-50 and US8K and outperforms comparable supervised methods results for musical auto-tagging on Magna-tag-a-tune.
2502.12033
The geometry of BERT
cs.LG
Transformer neural networks, particularly Bidirectional Encoder Representations from Transformers (BERT), have shown remarkable performance across various tasks such as classification, text summarization, and question answering. However, their internal mechanisms remain mathematically obscure, highlighting the need for greater explainability and interpretability. In this direction, this paper investigates the internal mechanisms of BERT proposing a novel perspective on the attention mechanism of BERT from a theoretical perspective. The analysis encompasses both local and global network behavior. At the local level, the concept of directionality of subspace selection as well as a comprehensive study of the patterns emerging from the self-attention matrix are presented. Additionally, this work explores the semantic content of the information stream through data distribution analysis and global statistical measures including the novel concept of cone index. A case study on the classification of SARS-CoV-2 variants using RNA which resulted in a very high accuracy has been selected in order to observe these concepts in an application. The insights gained from this analysis contribute to a deeper understanding of BERT's classification process, offering potential avenues for future architectural improvements in Transformer models and further analysis in the training process.
2502.12037
Information geometry of tempered stable processes
math.DG cs.IT math.IT math.PR
We find information geometry of tempered stable processes. Starting with the derivation of $\alpha$-divergence between two tempered stable processes, Fisher information matrices of tempered stable processes and $\alpha$-connections of their statistical manifolds are obtained. Additionally, we also provide statistical applications for the information geometry of tempered stable processes. Various tempered stable processes such as generalized tempered stable processes, classical tempered stable processes, and rapidly decreasing tempered stable processes are given as examples.
2502.12047
Quantum Byzantine Multiple Access Channels
cs.IT math.IT math.QA
In communication theory, attacks like eavesdropping or jamming are typically assumed to occur at the channel level, while communication parties are expected to follow established protocols. But what happens if one of the parties turns malicious? In this work, we investigate a compelling scenario: a multiple-access channel with two transmitters and one receiver, where one transmitter deviates from the protocol and acts dishonestly. To address this challenge, we introduce the Byzantine multiple-access classical-quantum channel and derive an achievable communication rate for this adversarial setting.
2502.12048
A Survey on Bridging EEG Signals and Generative AI: From Image and Text to Beyond
cs.AI cs.HC cs.LG
Integration of Brain-Computer Interfaces (BCIs) and Generative Artificial Intelligence (GenAI) has opened new frontiers in brain signal decoding, enabling assistive communication, neural representation learning, and multimodal integration. BCIs, particularly those leveraging Electroencephalography (EEG), provide a non-invasive means of translating neural activity into meaningful outputs. Recent advances in deep learning, including Generative Adversarial Networks (GANs) and Transformer-based Large Language Models (LLMs), have significantly improved EEG-based generation of images, text, and speech. This paper provides a literature review of the state-of-the-art in EEG-based multimodal generation, focusing on (i) EEG-to-image generation through GANs, Variational Autoencoders (VAEs), and Diffusion Models, and (ii) EEG-to-text generation leveraging Transformer based language models and contrastive learning methods. Additionally, we discuss the emerging domain of EEG-to-speech synthesis, an evolving multimodal frontier. We highlight key datasets, use cases, challenges, and EEG feature encoding methods that underpin generative approaches. By providing a structured overview of EEG-based generative AI, this survey aims to equip researchers and practitioners with insights to advance neural decoding, enhance assistive technologies, and expand the frontiers of brain-computer interaction.
2502.12049
Classifying the Stoichiometry of Virus-like Particles with Interpretable Machine Learning
cs.LG q-bio.BM q-bio.QM
Virus-like particles (VLPs) are valuable for vaccine development due to their immune-triggering properties. Understanding their stoichiometry, the number of protein subunits to form a VLP, is critical for vaccine optimisation. However, current experimental methods to determine stoichiometry are time-consuming and require highly purified proteins. To efficiently classify stoichiometry classes in proteins, we curate a new dataset and propose an interpretable, data-driven pipeline leveraging linear machine learning models. We also explore the impact of feature encoding on model performance and interpretability, as well as methods to identify key protein sequence features influencing classification. The evaluation of our pipeline demonstrates that it can classify stoichiometry while revealing protein features that possibly influence VLP assembly. The data and code used in this work are publicly available at https://github.com/Shef-AIRE/StoicIML.
2502.12050
SpeechT: Findings of the First Mentorship in Speech Translation
cs.CL cs.SD
This work presents the details and findings of the first mentorship in speech translation (SpeechT), which took place in December 2024 and January 2025. To fulfil the requirements of the mentorship, the participants engaged in key activities, including data preparation, modelling, and advanced research.
2502.12051
How to Upscale Neural Networks with Scaling Law? A Survey and Practical Guidelines
cs.CL cs.LG
Neural scaling laws have revolutionized the design and optimization of large-scale AI models by revealing predictable relationships between model size, dataset volume, and computational resources. Early research established power-law relationships in model performance, leading to compute-optimal scaling strategies. However, recent studies highlighted their limitations across architectures, modalities, and deployment contexts. Sparse models, mixture-of-experts, retrieval-augmented learning, and multimodal models often deviate from traditional scaling patterns. Moreover, scaling behaviors vary across domains such as vision, reinforcement learning, and fine-tuning, underscoring the need for more nuanced approaches. In this survey, we synthesize insights from over 50 studies, examining the theoretical foundations, empirical findings, and practical implications of scaling laws. We also explore key challenges, including data efficiency, inference scaling, and architecture-specific constraints, advocating for adaptive scaling strategies tailored to real-world applications. We suggest that while scaling laws provide a useful guide, they do not always generalize across all architectures and training strategies.
2502.12052
A Dual-Perspective NLG Meta-Evaluation Framework with Automatic Benchmark and Better Interpretability
cs.CL
In NLG meta-evaluation, evaluation metrics are typically assessed based on their consistency with humans. However, we identify some limitations in traditional NLG meta-evaluation approaches, such as issues in handling human ratings and ambiguous selections of correlation measures, which undermine the effectiveness of meta-evaluation. In this work, we propose a dual-perspective NLG meta-evaluation framework that focuses on different evaluation capabilities, thereby providing better interpretability. In addition, we introduce a method of automatically constructing the corresponding benchmarks without requiring new human annotations. Furthermore, we conduct experiments with 16 representative LLMs as the evaluators based on our proposed framework, comprehensively analyzing their evaluation performance from different perspectives.
2502.12054
PhysReason: A Comprehensive Benchmark towards Physics-Based Reasoning
cs.AI
Large language models demonstrate remarkable capabilities across various domains, especially mathematics and logic reasoning. However, current evaluations overlook physics-based reasoning - a complex task requiring physics theorems and constraints. We present PhysReason, a 1,200-problem benchmark comprising knowledge-based (25%) and reasoning-based (75%) problems, where the latter are divided into three difficulty levels (easy, medium, hard). Notably, problems require an average of 8.1 solution steps, with hard requiring 15.6, reflecting the complexity of physics-based reasoning. We propose the Physics Solution Auto Scoring Framework, incorporating efficient answer-level and comprehensive step-level evaluations. Top-performing models like Deepseek-R1, Gemini-2.0-Flash-Thinking, and o3-mini-high achieve less than 60% on answer-level evaluation, with performance dropping from knowledge questions (75.11%) to hard problems (31.95%). Through step-level evaluation, we identified four key bottlenecks: Physics Theorem Application, Physics Process Understanding, Calculation, and Physics Condition Analysis. These findings position PhysReason as a novel and comprehensive benchmark for evaluating physics-based reasoning capabilities in large language models. Our code and data will be published at https:/dxzxy12138.github.io/PhysReason.
2502.12055
Designing Role Vectors to Improve LLM Inference Behaviour
cs.CL
The influence of personas on Large Language Models (LLMs) has been widely studied, yet their direct impact on performance remains uncertain. This work explores a novel approach to guiding LLM behaviour through role vectors, an alternative to persona-based prompting. We construct 29 role vectors derived from model activations and evaluate their impact on benchmark performance across multiple domains. Our analysis investigates whether these vectors can effectively steer models toward domain-specific expertise. We measure two key interventions: (i) activation addition, which reinforces role-specific directions, and (ii) directional ablation, which removes them. Results on well-established benchmarks indicate that role vectors do, in fact, influence model behaviour, improving task performance in relevant domains while marginally affecting unrelated tasks. This, in turn, suggests that manipulating internal model representations has a greater impact on outcomes than persona-based prompting.
2502.12057
Culture is Not Trivia: Sociocultural Theory for Cultural NLP
cs.CL cs.CY
The field of cultural NLP has recently experienced rapid growth, driven by a pressing need to ensure that language technologies are effective and safe across a pluralistic user base. This work has largely progressed without a shared conception of culture, instead choosing to rely on a wide array of cultural proxies. However, this leads to a number of recurring limitations: coarse national boundaries fail to capture nuanced differences that lay within them, limited coverage restricts datasets to only a subset of usually highly-represented cultures, and a lack of dynamicity results in static cultural benchmarks that do not change as culture evolves. In this position paper, we argue that these methodological limitations are symptomatic of a theoretical gap. We draw on a well-developed theory of culture from sociocultural linguistics to fill this gap by 1) demonstrating in a case study how it can clarify methodological constraints and affordances, 2) offering theoretically-motivated paths forward to achieving cultural competence, and 3) arguing that localization is a more useful framing for the goals of much current work in cultural NLP.
2502.12058
A survey about perceptions of mobility to inform an agent-based simulator of subjective modal choice
cs.MA cs.CY
In order to adapt to the issues of climate change and public health, urban policies are trying to encourage soft mobility, but the share of the car remains significant. Beyond known constraints, we study here the impact of perception biases on individual choices. We designed a multi-criteria decision model, integrating the influence of habits and biases. We then conducted an online survey, which received 650 responses. We used these to calculate realistic mobility perception values, in order to initialise the environment and the population of a modal choice simulator, implemented in Netlogo. This allows us to visualize the adaptation of the modal distribution in reaction to the evolution of urban planning, depending on whether or not we activate biases and habits in individual reasoning. This is an extended and translated version of a demo paper published in French at JFSMA-JFMS 2024 "Un simulateur multi-agent de choix modal subjectif"
2502.12063
Low-Rank Thinning
stat.ML cs.LG math.OC math.ST stat.ME stat.TH
The goal in thinning is to summarize a dataset using a small set of representative points. Remarkably, sub-Gaussian thinning algorithms like Kernel Halving and Compress can match the quality of uniform subsampling while substantially reducing the number of summary points. However, existing guarantees cover only a restricted range of distributions and kernel-based quality measures and suffer from pessimistic dimension dependence. To address these deficiencies, we introduce a new low-rank analysis of sub-Gaussian thinning that applies to any distribution and any kernel, guaranteeing high-quality compression whenever the kernel or data matrix is approximately low-rank. To demonstrate the broad applicability of the techniques, we design practical sub-Gaussian thinning approaches that improve upon the best known guarantees for approximating attention in transformers, accelerating stochastic gradient training through reordering, and distinguishing distributions in near-linear time.
2502.12064
AI-generated Text Detection with a GLTR-based Approach
cs.CL cs.AI
The rise of LLMs (Large Language Models) has contributed to the improved performance and development of cutting-edge NLP applications. However, these can also pose risks when used maliciously, such as spreading fake news, harmful content, impersonating individuals, or facilitating school plagiarism, among others. This is because LLMs can generate high-quality texts, which are challenging to differentiate from those written by humans. GLTR, which stands for Giant Language Model Test Room and was developed jointly by the MIT-IBM Watson AI Lab and HarvardNLP, is a visual tool designed to help detect machine-generated texts based on GPT-2, that highlights the words in text depending on the probability that they were machine-generated. One limitation of GLTR is that the results it returns can sometimes be ambiguous and lead to confusion. This study aims to explore various ways to improve GLTR's effectiveness for detecting AI-generated texts within the context of the IberLef-AuTexTification 2023 shared task, in both English and Spanish languages. Experiment results show that our GLTR-based GPT-2 model overcomes the state-of-the-art models on the English dataset with a macro F1-score of 80.19%, except for the first ranking model (80.91%). However, for the Spanish dataset, we obtained a macro F1-score of 66.20%, which differs by 4.57% compared to the top-performing model.
2502.12065
Formalizing Complex Mathematical Statements with LLMs: A Study on Mathematical Definitions
cs.CL cs.FL
Thanks to their linguistic capabilities, LLMs offer an opportunity to bridge the gap between informal mathematics and formal languages through autoformalization. However, it is still unclear how well LLMs generalize to sophisticated and naturally occurring mathematical statements. To address this gap, we investigate the task of autoformalizing real-world mathematical definitions -- a critical component of mathematical discourse. Specifically, we introduce two novel resources for autoformalisation, collecting definitions from Wikipedia (Def_Wiki) and arXiv papers (Def_ArXiv). We then systematically evaluate a range of LLMs, analyzing their ability to formalize definitions into Isabelle/HOL. Furthermore, we investigate strategies to enhance LLMs' performance including refinement through external feedback from Proof Assistants, and formal definition grounding, where we guide LLMs through relevant contextual elements from formal mathematical libraries. Our findings reveal that definitions present a greater challenge compared to existing benchmarks, such as miniF2F. In particular, we found that LLMs still struggle with self-correction, and aligning with relevant mathematical libraries. At the same time, structured refinement methods and definition grounding strategies yield notable improvements of up to 16% on self-correction capabilities and 43% on the reduction of undefined errors, highlighting promising directions for enhancing LLM-based autoformalization in real-world scenarios.
2502.12066
CONSTRUCTA: Automating Commercial Construction Schedules in Fabrication Facilities with Large Language Models
cs.AI cs.LG cs.SE
Automating planning with LLMs presents transformative opportunities for traditional industries, yet remains underexplored. In commercial construction, the complexity of automated scheduling often requires manual intervention to ensure precision. We propose CONSTRUCTA, a novel framework leveraging LLMs to optimize construction schedules in complex projects like semiconductor fabrication. CONSTRUCTA addresses key challenges by: (1) integrating construction-specific knowledge through static RAG; (2) employing context-sampling techniques inspired by architectural expertise to provide relevant input; and (3) deploying Construction DPO to align schedules with expert preferences using RLHF. Experiments on proprietary data demonstrate performance improvements of +42.3% in missing value prediction, +79.1% in dependency analysis, and +28.9% in automated planning compared to baseline methods, showcasing its potential to revolutionize construction workflows and inspire domain-specific LLM advancements.
2502.12067
TokenSkip: Controllable Chain-of-Thought Compression in LLMs
cs.CL cs.AI
Chain-of-Thought (CoT) has been proven effective in enhancing the reasoning capabilities of large language models (LLMs). Recent advancements, such as OpenAI's o1 and DeepSeek-R1, suggest that scaling up the length of CoT sequences during inference could further boost LLM reasoning performance. However, due to the autoregressive nature of LLM decoding, longer CoT outputs lead to a linear increase in inference latency, adversely affecting user experience, particularly when the CoT exceeds 10,000 tokens. To address this limitation, we analyze the semantic importance of tokens within CoT outputs and reveal that their contributions to reasoning vary. Building on this insight, we propose TokenSkip, a simple yet effective approach that enables LLMs to selectively skip less important tokens, allowing for controllable CoT compression. Extensive experiments across various models and tasks demonstrate the effectiveness of TokenSkip in reducing CoT token usage while preserving strong reasoning performance. Notably, when applied to Qwen2.5-14B-Instruct, TokenSkip reduces reasoning tokens by 40% (from 313 to 181) on GSM8K, with less than a 0.4% performance drop.
2502.12073
Can LLMs Simulate Social Media Engagement? A Study on Action-Guided Response Generation
cs.CL
Social media enables dynamic user engagement with trending topics, and recent research has explored the potential of large language models (LLMs) for response generation. While some studies investigate LLMs as agents for simulating user behavior on social media, their focus remains on practical viability and scalability rather than a deeper understanding of how well LLM aligns with human behavior. This paper analyzes LLMs' ability to simulate social media engagement through action guided response generation, where a model first predicts a user's most likely engagement action-retweet, quote, or rewrite-towards a trending post before generating a personalized response conditioned on the predicted action. We benchmark GPT-4o-mini, O1-mini, and DeepSeek-R1 in social media engagement simulation regarding a major societal event discussed on X. Our findings reveal that zero-shot LLMs underperform BERT in action prediction, while few-shot prompting initially degrades the prediction accuracy of LLMs with limited examples. However, in response generation, few-shot LLMs achieve stronger semantic alignment with ground truth posts.
2502.12080
HumanGif: Single-View Human Diffusion with Generative Prior
cs.CV
While previous single-view-based 3D human reconstruction methods made significant progress in novel view synthesis, it remains a challenge to synthesize both view-consistent and pose-consistent results for animatable human avatars from a single image input. Motivated by the success of 2D character animation, we propose <strong>HumanGif</strong>, a single-view human diffusion model with generative prior. Specifically, we formulate the single-view-based 3D human novel view and pose synthesis as a single-view-conditioned human diffusion process, utilizing generative priors from foundational diffusion models. To ensure fine-grained and consistent novel view and pose synthesis, we introduce a Human NeRF module in HumanGif to learn spatially aligned features from the input image, implicitly capturing the relative camera and human pose transformation. Furthermore, we introduce an image-level loss during optimization to bridge the gap between latent and image spaces in diffusion models. Extensive experiments on RenderPeople and DNA-Rendering datasets demonstrate that HumanGif achieves the best perceptual performance, with better generalizability for novel view and pose synthesis.
2502.12081
Unhackable Temporal Rewarding for Scalable Video MLLMs
cs.CV cs.CL
In the pursuit of superior video-processing MLLMs, we have encountered a perplexing paradox: the "anti-scaling law", where more data and larger models lead to worse performance. This study unmasks the culprit: "temporal hacking", a phenomenon where models shortcut by fixating on select frames, missing the full video narrative. In this work, we systematically establish a comprehensive theory of temporal hacking, defining it from a reinforcement learning perspective, introducing the Temporal Perplexity (TPL) score to assess this misalignment, and proposing the Unhackable Temporal Rewarding (UTR) framework to mitigate the temporal hacking. Both theoretically and empirically, TPL proves to be a reliable indicator of temporal modeling quality, correlating strongly with frame activation patterns. Extensive experiments reveal that UTR not only counters temporal hacking but significantly elevates video comprehension capabilities. This work not only advances video-AI systems but also illuminates the critical importance of aligning proxy rewards with true objectives in MLLM development.
2502.12082
AdaSplash: Adaptive Sparse Flash Attention
cs.CL cs.LG
The computational cost of softmax-based attention in transformers limits their applicability to long-context tasks. Adaptive sparsity, of which $\alpha$-entmax attention is an example, offers a flexible data-dependent alternative, but existing implementations are inefficient and do not leverage the sparsity to obtain runtime and memory gains. In this work, we propose AdaSplash, which combines the efficiency of GPU-optimized algorithms with the sparsity benefits of $\alpha$-entmax. We first introduce a hybrid Halley-bisection algorithm, resulting in a 7-fold reduction in the number of iterations needed to compute the $\alpha$-entmax transformation. Then, we implement custom Triton kernels to efficiently handle adaptive sparsity. Experiments with RoBERTa and ModernBERT for text classification and single-vector retrieval, along with GPT-2 for language modeling, show that our method achieves substantial improvements in runtime and memory efficiency compared to existing $\alpha$-entmax implementations. It approaches -- and in some cases surpasses -- the efficiency of highly optimized softmax implementations like FlashAttention-2, enabling long-context training while maintaining strong task performance.
2502.12084
VLM$^2$-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual Cues
cs.CL
Visually linking matching cues is a crucial ability in daily life, such as identifying the same person in multiple photos based on their cues, even without knowing who they are. Despite the extensive knowledge that vision-language models (VLMs) possess, it remains largely unexplored whether they are capable of performing this fundamental task. To address this, we introduce VLM$^2$-Bench, a benchmark designed to assess whether VLMs can Visually Link Matching cues, with 9 subtasks and over 3,000 test cases. Comprehensive evaluation across eight open-source VLMs and GPT-4o, along with further analysis of various language-side and vision-side prompting methods, leads to a total of eight key findings. We identify critical challenges in models' ability to link visual cues, highlighting a significant performance gap where even GPT-4o lags 34.80% behind humans. Based on these insights, we advocate for (i) enhancing core visual capabilities to improve adaptability and reduce reliance on prior knowledge, (ii) establishing clearer principles for integrating language-based reasoning in vision-centric tasks to prevent unnecessary biases, and (iii) shifting vision-text training paradigms toward fostering models' ability to independently structure and infer relationships among visual cues.
2502.12085
APB: Accelerating Distributed Long-Context Inference by Passing Compressed Context Blocks across GPUs
cs.LG cs.CL
While long-context inference is crucial for advancing large language model (LLM) applications, its prefill speed remains a significant bottleneck. Current approaches, including sequence parallelism strategies and compute reduction through approximate attention mechanisms, still fall short of delivering optimal inference efficiency. This hinders scaling the inputs to longer sequences and processing long-context queries in a timely manner. To address this, we introduce APB, an efficient long-context inference framework that leverages multi-host approximate attention to enhance prefill speed by reducing compute and enhancing parallelism simultaneously. APB introduces a communication mechanism for essential key-value pairs within a sequence parallelism framework, enabling a faster inference speed while maintaining task performance. We implement APB by incorporating a tailored FlashAttn kernel alongside optimized distribution strategies, supporting diverse models and parallelism configurations. APB achieves speedups of up to 9.2x, 4.2x, and 1.6x compared with FlashAttn, RingAttn, and StarAttn, respectively, without any observable task performance degradation. We provide the implementation and experiment code of APB in https://github.com/thunlp/APB.
2502.12086
Unifying Explainable Anomaly Detection and Root Cause Analysis in Dynamical Systems
cs.LG stat.ML
Dynamical systems, prevalent in various scientific and engineering domains, are susceptible to anomalies that can significantly impact their performance and reliability. This paper addresses the critical challenges of anomaly detection, root cause localization, and anomaly type classification in dynamical systems governed by ordinary differential equations (ODEs). We define two categories of anomalies: cyber anomalies, which propagate through interconnected variables, and measurement anomalies, which remain localized to individual variables. To address these challenges, we propose the Interpretable Causality Ordinary Differential Equation (ICODE) Networks, a model-intrinsic explainable learning framework. ICODE leverages Neural ODEs for anomaly detection while employing causality inference through an explanation channel to perform root cause analysis (RCA), elucidating why specific time periods are flagged as anomalous. ICODE is designed to simultaneously perform anomaly detection, RCA, and anomaly type classification within a single, interpretable framework. Our approach is grounded in the hypothesis that anomalies alter the underlying ODEs of the system, manifesting as changes in causal relationships between variables. We provide a theoretical analysis of how perturbations in learned model parameters can be utilized to identify anomalies and their root causes in time series data. Comprehensive experimental evaluations demonstrate the efficacy of ICODE across various dynamical systems, showcasing its ability to accurately detect anomalies, classify their types, and pinpoint their origins.
2502.12088
Meta-Statistical Learning: Supervised Learning of Statistical Inference
cs.LG cs.AI
This work demonstrates that the tools and principles driving the success of large language models (LLMs) can be repurposed to tackle distribution-level tasks, where the goal is to predict properties of the data-generating distribution rather than labels for individual datapoints. These tasks encompass statistical inference problems such as parameter estimation, hypothesis testing, or mutual information estimation. Framing these tasks within traditional machine learning pipelines is challenging, as supervision is typically tied to individual datapoint. We propose meta-statistical learning, a framework inspired by multi-instance learning that reformulates statistical inference tasks as supervised learning problems. In this approach, entire datasets are treated as single inputs to neural networks, which predict distribution-level parameters. Transformer-based architectures, without positional encoding, provide a natural fit due to their permutation-invariance properties. By training on large-scale synthetic datasets, meta-statistical models can leverage the scalability and optimization infrastructure of Transformer-based LLMs. We demonstrate the framework's versatility with applications in hypothesis testing and mutual information estimation, showing strong performance, particularly for small datasets where traditional neural methods struggle.
2502.12089
How compositional generalization and creativity improve as diffusion models are trained
stat.ML cs.LG
Natural data is often organized as a hierarchical composition of features. How many samples do generative models need to learn the composition rules, so as to produce a combinatorial number of novel data? What signal in the data is exploited to learn? We investigate these questions both theoretically and empirically. Theoretically, we consider diffusion models trained on simple probabilistic context-free grammars - tree-like graphical models used to represent the structure of data such as language and images. We demonstrate that diffusion models learn compositional rules with the sample complexity required for clustering features with statistically similar context, a process similar to the word2vec algorithm. However, this clustering emerges hierarchically: higher-level, more abstract features associated with longer contexts require more data to be identified. This mechanism leads to a sample complexity that scales polynomially with the said context size. As a result, diffusion models trained on intermediate dataset size generate data coherent up to a certain scale, but that lacks global coherence. We test these predictions in different domains, and find remarkable agreement: both generated texts and images achieve progressively larger coherence lengths as the training time or dataset size grows. We discuss connections between the hierarchical clustering mechanism we introduce here and the renormalization group in physics.
2502.12093
WeVibe: Weight Change Estimation Through Audio-Induced Shelf Vibrations In Autonomous Stores
eess.SP cs.SY eess.SY
Weight change estimation is crucial in various applications, particularly for detecting pick-up and put-back actions when people interact with the shelf while shopping in autonomous stores. Moreover, accurate weight change estimation allows autonomous stores to automatically identify items being picked up or put back, ensuring precise cost estimation. However, the conventional approach of estimating weight changes requires specialized weight-sensing shelves, which are densely deployed weight scales, incurring intensive sensor consumption and high costs. Prior works explored the vibration-based weight sensing method, but they failed when the location of weight change varies. In response to these limitations, we made the following contributions: (1) We propose WeVibe, a first item weight change estimation system through active shelf vibration sensing. The main intuition of the system is that the weight placed on the shelf influences the dynamic vibration response of the shelf, thus altering the shelf vibration patterns. (2) We model a physics-informed relationship between the shelf vibration response and item weight across multiple locations on the shelf based on structural dynamics theory. This relationship is linear and allows easy training of a weight estimation model at a new location without heavy data collection. (3) We evaluate our system on a gondola shelf organized as the real-store settings. WeVibe achieved a mean absolute error down to 38.07g and a standard deviation of 31.2g with one sensor and 10% samples from three weight classes on estimating weight change from 0g to 450g, which can be leveraged for differentiating items with more than 100g differences.
2502.12094
A Study on Leveraging Search and Self-Feedback for Agent Reasoning
cs.AI cs.CL
Recent works have demonstrated that incorporating search during inference can significantly improve reasoning capabilities of language agents. Some approaches may make use of the ground truth or rely on model's own generated feedback. The search algorithm uses this feedback to then produce values that will update its criterion for exploring and exploiting various reasoning paths. In this study, we investigate how search and model's self-feedback can be leveraged for reasoning tasks. First, we explore differences in ground-truth feedback and self-feedback during search for math reasoning. Second, we observe limitations in applying search techniques to more complex tasks like tool-calling and design domain-specific approaches to address these gaps. Our experiments reveal challenges related to generalization when solely relying on self-feedback during search. For search to work effectively, either access to the ground-truth is needed or feedback mechanisms need to be carefully designed for the specific task.
2502.12095
Descriminative-Generative Custom Tokens for Vision-Language Models
cs.CV
This paper explores the possibility of learning custom tokens for representing new concepts in Vision-Language Models (VLMs). Our aim is to learn tokens that can be effective for both discriminative and generative tasks while composing well with words to form new input queries. The targeted concept is specified in terms of a small set of images and a parent concept described using text. We operate on CLIP text features and propose to use a combination of a textual inversion loss and a classification loss to ensure that text features of the learned token are aligned with image features of the concept in the CLIP embedding space. We restrict the learned token to a low-dimensional subspace spanned by tokens for attributes that are appropriate for the given super-class. These modifications improve the quality of compositions of the learned token with natural language for generating new scenes. Further, we show that learned custom tokens can be used to form queries for text-to-image retrieval task, and also have the important benefit that composite queries can be visualized to ensure that the desired concept is faithfully encoded. Based on this, we introduce the method of Generation Aided Image Retrieval, where the query is modified at inference time to better suit the search intent. On the DeepFashion2 dataset, our method improves Mean Reciprocal Retrieval (MRR) over relevant baselines by 7%.
2502.12096
Token Communications: A Unified Framework for Cross-modal Context-aware Semantic Communications
cs.IT cs.CV cs.MM eess.SP math.IT
In this paper, we introduce token communications (TokCom), a unified framework to leverage cross-modal context information in generative semantic communications (GenSC). TokCom is a new paradigm, motivated by the recent success of generative foundation models and multimodal large language models (GFM/MLLMs), where the communication units are tokens, enabling efficient transformer-based token processing at the transmitter and receiver. In this paper, we introduce the potential opportunities and challenges of leveraging context in GenSC, explore how to integrate GFM/MLLMs-based token processing into semantic communication systems to leverage cross-modal context effectively, present the key principles for efficient TokCom at various layers in future wireless networks. We demonstrate the corresponding TokCom benefits in a GenSC setup for image, leveraging cross-modal context information, which increases the bandwidth efficiency by 70.8% with negligible loss of semantic/perceptual quality. Finally, the potential research directions are identified to facilitate adoption of TokCom in future wireless networks.
2502.12098
Bandwidth-Adaptive Spatiotemporal Correspondence Identification for Collaborative Perception
cs.RO
Correspondence identification (CoID) is an essential capability in multi-robot collaborative perception, which enables a group of robots to consistently refer to the same objects within their respective fields of view. In real-world applications, such as connected autonomous driving, vehicles face challenges in directly sharing raw observations due to limited communication bandwidth. In order to address this challenge, we propose a novel approach for bandwidth-adaptive spatiotemporal CoID in collaborative perception. This approach allows robots to progressively select partial spatiotemporal observations and share with others, while adapting to communication constraints that dynamically change over time. We evaluate our approach across various scenarios in connected autonomous driving simulations. Experimental results validate that our approach enables CoID and adapts to dynamic communication bandwidth changes. In addition, our approach achieves 8%-56% overall improvements in terms of covisible object retrieval for CoID and data sharing efficiency, which outperforms previous techniques and achieves the state-of-the-art performance. More information is available at: https://gaopeng5.github.io/acoid.
2502.12102
Relational Norms for Human-AI Cooperation
cs.AI cs.ET
How we should design and interact with social artificial intelligence depends on the socio-relational role the AI is meant to emulate or occupy. In human society, relationships such as teacher-student, parent-child, neighbors, siblings, or employer-employee are governed by specific norms that prescribe or proscribe cooperative functions including hierarchy, care, transaction, and mating. These norms shape our judgments of what is appropriate for each partner. For example, workplace norms may allow a boss to give orders to an employee, but not vice versa, reflecting hierarchical and transactional expectations. As AI agents and chatbots powered by large language models are increasingly designed to serve roles analogous to human positions - such as assistant, mental health provider, tutor, or romantic partner - it is imperative to examine whether and how human relational norms should extend to human-AI interactions. Our analysis explores how differences between AI systems and humans, such as the absence of conscious experience and immunity to fatigue, may affect an AI's capacity to fulfill relationship-specific functions and adhere to corresponding norms. This analysis, which is a collaborative effort by philosophers, psychologists, relationship scientists, ethicists, legal experts, and AI researchers, carries important implications for AI systems design, user behavior, and regulation. While we accept that AI systems can offer significant benefits such as increased availability and consistency in certain socio-relational roles, they also risk fostering unhealthy dependencies or unrealistic expectations that could spill over into human-human relationships. We propose that understanding and thoughtfully shaping (or implementing) suitable human-AI relational norms will be crucial for ensuring that human-AI interactions are ethical, trustworthy, and favorable to human well-being.
2502.12108
Using the Path of Least Resistance to Explain Deep Networks
cs.LG cs.AI stat.ML
Integrated Gradients (IG), a widely used axiomatic path-based attribution method, assigns importance scores to input features by integrating model gradients along a straight path from a baseline to the input. While effective in some cases, we show that straight paths can lead to flawed attributions. In this paper, we identify the cause of these misattributions and propose an alternative approach that treats the input space as a Riemannian manifold, computing attributions by integrating gradients along geodesics. We call this method Geodesic Integrated Gradients (GIG). To approximate geodesic paths, we introduce two techniques: a k-Nearest Neighbours-based approach for smaller models and a Stochastic Variational Inference-based method for larger ones. Additionally, we propose a new axiom, Strong Completeness, extending the axioms satisfied by IG. We show that this property is desirable for attribution methods and that GIG is the only method that satisfies it. Through experiments on both synthetic and real-world data, we demonstrate that GIG outperforms existing explainability methods, including IG.
2502.12109
Personality Structured Interview for Large Language Model Simulation in Personality Research
cs.CL cs.AI
Although psychometrics researchers have recently explored the use of large language models (LLMs) as proxies for human participants, LLMs often fail to generate heterogeneous data with human-like diversity, which diminishes their value in advancing social science research. To address these challenges, we explored the potential of the theory-informed Personality Structured Interview (PSI) as a tool for simulating human responses in personality research. In this approach, the simulation is grounded in nuanced real-human interview transcripts that target the personality construct of interest. We have provided a growing set of 357 structured interview transcripts from a representative sample, each containing an individual's response to 32 open-ended questions carefully designed to gather theory-based personality evidence. Additionally, grounded in psychometric research, we have summarized an evaluation framework to systematically validate LLM-generated psychometric data. Results from three experiments demonstrate that well-designed structured interviews could improve human-like heterogeneity in LLM-simulated personality data and predict personality-related behavioral outcomes (i.e., organizational citizenship behaviors and counterproductive work behavior). We further discuss the role of theory-informed structured interviews in LLM-based simulation and outline a general framework for designing structured interviews to simulate human-like data for psychometric research.
2502.12110
A-MEM: Agentic Memory for LLM Agents
cs.CL cs.HC
While large language model (LLM) agents can effectively use external tools for complex real-world tasks, they require memory systems to leverage historical experiences. Current memory systems enable basic storage and retrieval but lack sophisticated memory organization, despite recent attempts to incorporate graph databases. Moreover, these systems' fixed operations and structures limit their adaptability across diverse tasks. To address this limitation, this paper proposes a novel agentic memory system for LLM agents that can dynamically organize memories in an agentic way. Following the basic principles of the Zettelkasten method, we designed our memory system to create interconnected knowledge networks through dynamic indexing and linking. When a new memory is added, we generate a comprehensive note containing multiple structured attributes, including contextual descriptions, keywords, and tags. The system then analyzes historical memories to identify relevant connections, establishing links where meaningful similarities exist. Additionally, this process enables memory evolution - as new memories are integrated, they can trigger updates to the contextual representations and attributes of existing historical memories, allowing the memory network to continuously refine its understanding. Our approach combines the structured organization principles of Zettelkasten with the flexibility of agent-driven decision making, allowing for more adaptive and context-aware memory management. Empirical experiments on six foundation models show superior improvement against existing SOTA baselines. The source code is available at https://github.com/WujiangXu/AgenticMemory.
2502.12113
A Monocular Event-Camera Motion Capture System
cs.RO cs.CV
Motion capture systems are a widespread tool in research to record ground-truth poses of objects. Commercial systems use reflective markers attached to the object and then triangulate pose of the object from multiple camera views. Consequently, the object must be visible to multiple cameras which makes such multi-view motion capture systems unsuited for deployments in narrow, confined spaces (e.g. ballast tanks of ships). In this technical report we describe a monocular event-camera motion capture system which overcomes this limitation and is ideally suited for narrow spaces. Instead of passive markers it relies on active, blinking LED markers such that each marker can be uniquely identified from the blinking frequency. The markers are placed at known locations on the tracking object. We then solve the PnP (perspective-n-points) problem to obtain the position and orientation of the object. The developed system has millimeter accuracy, millisecond latency and we demonstrate that its state estimate can be used to fly a small, agile quadrotor.
2502.12115
SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?
cs.LG cs.SE
We introduce SWE-Lancer, a benchmark of over 1,400 freelance software engineering tasks from Upwork, valued at \$1 million USD total in real-world payouts. SWE-Lancer encompasses both independent engineering tasks--ranging from \$50 bug fixes to \$32,000 feature implementations--and managerial tasks, where models choose between technical implementation proposals. Independent tasks are graded with end-to-end tests triple-verified by experienced software engineers, while managerial decisions are assessed against the choices of the original hired engineering managers. We evaluate model performance and find that frontier models are still unable to solve the majority of tasks. To facilitate future research, we open-source a unified Docker image and a public evaluation split, SWE-Lancer Diamond (https://github.com/openai/SWELancer-Benchmark). By mapping model performance to monetary value, we hope SWE-Lancer enables greater research into the economic impact of AI model development.
2502.12118
Scaling Test-Time Compute Without Verification or RL is Suboptimal
cs.LG cs.CL
Despite substantial advances in scaling test-time compute, an ongoing debate in the community is how it should be scaled up to enable continued and efficient improvements with scaling. There are largely two approaches: first, distilling successful search or thinking traces; and second, using verification (e.g., 0/1 outcome rewards, reward models, or verifiers) to guide reinforcement learning (RL) and search algorithms. In this paper, we prove that finetuning LLMs with verifier-based (VB) methods based on RL or search is far superior to verifier-free (VF) approaches based on distilling or cloning search traces, given a fixed amount of compute/data budget. Further, we show that as we scale test-time compute (measured as the output token length) and training data, suboptimality of VF methods scales poorly compared to VB when the base pre-trained LLM presents a heterogeneous distribution over correct solution traces (e.g., different lengths, styles, etc.) and admits a non-sharp distribution over rewards on traces sampled from it. We formalize this condition using anti-concentration [Erd\H{o}s, 1945]. This implies a stronger result that VB methods scale better asymptotically, with the performance gap between VB and VF methods widening as test-time budget grows. We corroborate our theory empirically on both didactic and math reasoning problems with 3/8/32B-sized pre-trained LLMs, where we find verification is crucial for scaling test-time compute.
2502.12119
PRISM: Self-Pruning Intrinsic Selection Method for Training-Free Multimodal Data Selection
cs.CV cs.AI cs.CL
Visual instruction tuning refines pre-trained Multimodal Large Language Models (MLLMs) to enhance their real-world task performance. However, the rapid expansion of visual instruction datasets introduces significant data redundancy, leading to excessive computational costs. Existing data selection methods predominantly rely on proxy models or loss-based metrics, both of which impose substantial computational overheads due to the necessity of model inference and backpropagation. To address this challenge, we propose PRISM, a novel training-free approach for efficient multimodal data selection. Unlike existing methods, PRISM eliminates the reliance on proxy models, warm-up pretraining, and gradient-based optimization. Instead, it leverages Pearson correlation analysis to quantify the intrinsic visual encoding properties of MLLMs, computing a task-specific correlation score to identify high-value instances. This not only enbles data-efficient selection,but maintains the original performance. Empirical evaluations across multiple MLLMs demonstrate that PRISM reduces the overall time required for visual instruction tuning and data selection to just 30% of conventional methods, while surpassing fully fine-tuned models across eight multimodal and three language understanding benchmarks, achieving a 101.7% relative improvement in final performance.
2502.12120
LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws
cs.LG cs.AI cs.CL
Scaling laws guide the development of large language models (LLMs) by offering estimates for the optimal balance of model size, tokens, and compute. More recently, loss-to-loss scaling laws that relate losses across pretraining datasets and downstream tasks have emerged as a powerful tool for understanding and improving LLM performance. In this work, we investigate which factors most strongly influence loss-to-loss scaling. Our experiments reveal that the pretraining data and tokenizer determine the scaling trend. In contrast, model size, optimization hyperparameters, and even significant architectural differences, such as between transformer-based models like Llama and state-space models like Mamba, have limited impact. Consequently, practitioners should carefully curate suitable pretraining datasets for optimal downstream performance, while architectures and other settings can be freely optimized for training efficiency.
2502.12122
Minimal Ranks, Maximum Confidence: Parameter-efficient Uncertainty Quantification for LoRA
cs.LG
Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning of large language models by decomposing weight updates into low-rank matrices, significantly reducing storage and computational overhead. While effective, standard LoRA lacks mechanisms for uncertainty quantification, leading to overconfident and poorly calibrated models. Bayesian variants of LoRA address this limitation, but at the cost of a significantly increased number of trainable parameters, partially offsetting the original efficiency gains. Additionally, these models are harder to train and may suffer from unstable convergence. In this work, we propose a novel parameter-efficient Bayesian LoRA, demonstrating that effective uncertainty quantification can be achieved in very low-dimensional parameter spaces. The proposed method achieves strong performance with improved calibration and generalization while maintaining computational efficiency. Our empirical findings show that, with the appropriate projection of the weight space: (1) uncertainty can be effectively modeled in a low-dimensional space, and (2) weight covariances exhibit low ranks.
2502.12123
On the Query Complexity of Verifier-Assisted Language Generation
cs.CL cs.LG
Recently, a plethora of works have proposed inference-time algorithms (e.g. best-of-n), which incorporate verifiers to assist the generation process. Their quality-efficiency trade-offs have been empirically benchmarked on a variety of constrained generation tasks, but the algorithmic design landscape is still largely poorly understood. In this paper, we develop a mathematical framework for reasoning about constrained generation using a pre-trained language model generator oracle and a process verifier--which can decide whether a prefix can be extended to a string which satisfies the constraints of choice. We show that even in very simple settings, access to a verifier can render an intractable problem (information-theoretically or computationally) to a tractable one. In fact, we show even simple algorithms, like tokenwise rejection sampling, can enjoy significant benefits from access to a verifier. Empirically, we show that a natural modification of tokenwise rejection sampling, in which the sampler is allowed to "backtrack" (i.e., erase the final few generated tokens) has robust and substantive benefits over natural baselines (e.g. (blockwise) rejection sampling, nucleus sampling)--both in terms of computational efficiency, accuracy and diversity.
2502.12124
RA-MTR: A Retrieval Augmented Multi-Task Reader based Approach for Inspirational Quote Extraction from Long Documents
cs.CL
Inspirational quotes from famous individuals are often used to convey thoughts in news articles, essays, and everyday conversations. In this paper, we propose a novel context-based quote extraction system that aims to extract the most relevant quote from a long text. We formulate this quote extraction as an open domain question answering problem first by employing a vector-store based retriever and then applying a multi-task reader. We curate three context-based quote extraction datasets and introduce a novel multi-task framework RA-MTR that improves the state-of-the-art performance, achieving a maximum improvement of 5.08% in BoW F1-score.
2502.12125
Hypernym Bias: Unraveling Deep Classifier Training Dynamics through the Lens of Class Hierarchy
cs.AI cs.LG
We investigate the training dynamics of deep classifiers by examining how hierarchical relationships between classes evolve during training. Through extensive experiments, we argue that the learning process in classification problems can be understood through the lens of label clustering. Specifically, we observe that networks tend to distinguish higher-level (hypernym) categories in the early stages of training, and learn more specific (hyponym) categories later. We introduce a novel framework to track the evolution of the feature manifold during training, revealing how the hierarchy of class relations emerges and refines across the network layers. Our analysis demonstrates that the learned representations closely align with the semantic structure of the dataset, providing a quantitative description of the clustering process. Notably, we show that in the hypernym label space, certain properties of neural collapse appear earlier than in the hyponym label space, helping to bridge the gap between the initial and terminal phases of learning. We believe our findings offer new insights into the mechanisms driving hierarchical learning in deep networks, paving the way for future advancements in understanding deep learning dynamics.