×
![]()
arXiv:2504.10885v1 Announce Type: cross
Abstract: Large Multimodal Models (LMMs) have demonstrated impressive capabilities across a wide range of multimodal tasks, achieving ever-increasing performance on various evaluation benchmarks. However, existing benchmarks are typically static and often overlap with pre-training datasets, leading to fixed complexity constraints and substantial data contamination issues. Meanwhile, manually annotated datasets are labor-intensive, time-consuming, and subject to human bias and inconsistency, leading to reliability and reproducibility issues. To address these problems, we propose a fully dynamic multimodal evaluation framework, named Open-ended Visual Puzzle Generation (OVPG), which aims to generate fresh, diverse, and verifiable evaluation data automatically in puzzle-solving tasks. Specifically, the OVPG pipeline consists of a raw material sampling module, a visual content generation module, and a puzzle rule design module, which ensures that each evaluation instance is primitive, highly randomized, and uniquely solvable, enabling continual adaptation to the evolving capabilities of LMMs. Built upon OVPG, we construct PuzzleBench, a dynamic and scalable benchmark comprising 11,840 VQA samples. It features six carefully designed puzzle tasks targeting three core LMM competencies, visual recognition, logical reasoning, and context understanding. PuzzleBench differs from static benchmarks that quickly become outdated. It enables ongoing dataset refreshing through OVPG and a rich set of open-ended puzzle designs, allowing seamless adaptation to the evolving capabilities of LMMs.
arXiv:2409.19552v3 Announce Type: replace-cross
Abstract: X-ray absorption spectroscopy (XAS) is a powerful characterization technique for probing the local chemical environment of absorbing atoms. However, analyzing XAS data presents significant challenges, often requiring extensive, computationally intensive simulations, as well as significant domain expertise. These limitations hinder the development of fast, robust XAS analysis pipelines that are essential in high-throughput studies and for autonomous experimentation. We address these challenges with OmniXAS, a framework that contains a suite of transfer learning approaches for XAS prediction, each contributing to improved accuracy and efficiency, as demonstrated on K-edge spectra database covering eight 3d transition metals (Ti-Cu). The OmniXAS framework is built upon three distinct strategies. First, we use M3GNet to derive latent representations of the local chemical environment of absorption sites as input for XAS prediction, achieving up to order-of-magnitude improvements over conventional featurization techniques. Second, we employ a hierarchical transfer learning strategy, training a universal multi-task model across elements before fine-tuning for element-specific predictions. Models based on this cascaded approach after element-wise fine-tuning outperform element-specific models by up to 69%. Third, we implement cross-fidelity transfer learning, adapting a universal model to predict spectra generated by simulation of a different fidelity with a higher computational cost. This approach improves prediction accuracy by up to 11% over models trained on the target fidelity alone. Our approach boosts the throughput of XAS modeling by orders of magnitude versus first-principles simulations and is extendable to XAS prediction for a broader range of elements. This transfer learning framework is generalizable to enhance deep-learning models that target other properties in materials research.
arXiv:2504.10781v1 Announce Type: cross
Abstract: The classical limit of quantum mechanics, formally investigated through frameworks like strict deformation quantization, remains a profound area of inquiry in the philosophy of physics. This paper explores a computational approach employing a neural network to emulate the emergence of classical behavior from the quantum harmonic oscillator as Planck's constant $\hbar$ approaches zero. We develop and train a neural network architecture to learn the mapping from initial expectation values and $\hbar$ to the time evolution of the expectation value of position. By analyzing the network's predictions across different regimes of hbar, we aim to provide computational insights into the nature of the quantum-classical transition. This work demonstrates the potential of machine learning as a complementary tool for exploring foundational questions in quantum mechanics and its classical limit.
arXiv:2504.10900v1 Announce Type: cross
Abstract: Foundation models have achieved remarkable success across diverse machine-learning domains through large-scale pretraining on large, diverse datasets. However, pretraining on such datasets introduces significant challenges due to substantial mismatches in data distributions, a problem particularly pronounced with time series data. In this paper, we tackle this issue by proposing a domain-aware adaptive normalization strategy within the Transformer architecture. Specifically, we replace the traditional LayerNorm with a prototype-guided dynamic normalization mechanism (ProtoNorm), where learned prototypes encapsulate distinct data distributions, and sample-to-prototype affinity determines the appropriate normalization layer. This mechanism effectively captures the heterogeneity of time series characteristics, aligning pretrained representations with downstream tasks. Through comprehensive empirical evaluation, we demonstrate that our method significantly outperforms conventional pretraining techniques across both classification and forecasting tasks, while effectively mitigating the adverse effects of distribution shifts during pretraining. Incorporating ProtoNorm is as simple as replacing a single line of code. Extensive experiments on diverse real-world time series benchmarks validate the robustness and generalizability of our approach, advancing the development of more versatile time series foundation models.
arXiv:2503.05628v2 Announce Type: replace-cross
Abstract: Rapid advances in AI are beginning to reshape national security. Destabilizing AI developments could rupture the balance of power and raise the odds of great-power conflict, while widespread proliferation of capable AI hackers and virologists would lower barriers for rogue actors to cause catastrophe. Superintelligence -- AI vastly better than humans at nearly all cognitive tasks -- is now anticipated by AI researchers. Just as nations once developed nuclear strategies to secure their survival, we now need a coherent superintelligence strategy to navigate a new period of transformative change. We introduce the concept of Mutual Assured AI Malfunction (MAIM): a deterrence regime resembling nuclear mutual assured destruction (MAD) where any state's aggressive bid for unilateral AI dominance is met with preventive sabotage by rivals. Given the relative ease of sabotaging a destabilizing AI project -- through interventions ranging from covert cyberattacks to potential kinetic strikes on datacenters -- MAIM already describes the strategic picture AI superpowers find themselves in. Alongside this, states can increase their competitiveness by bolstering their economies and militaries through AI, and they can engage in nonproliferation to rogue actors to keep weaponizable AI capabilities out of their hands. Taken together, the three-part framework of deterrence, nonproliferation, and competitiveness outlines a robust strategy to superintelligence in the years ahead.
arXiv:2504.10538v1 Announce Type: cross
Abstract: Session-based recommendation (SBR) predicts the next item based on anonymous sessions. Traditional SBR explores user intents based on ID collaborations or auxiliary content. To further alleviate data sparsity and cold-start issues, recent Multimodal SBR (MSBR) methods utilize simplistic pre-trained models for modality learning but have limitations in semantic richness. Considering semantic reasoning abilities of Large Language Models (LLM), we focus on the LLM-enhanced MSBR scenario in this paper, which leverages LLM cognition for comprehensive multimodal representation generation, to enhance downstream MSBR. Tackling this problem faces two challenges: i) how to obtain LLM cognition on both transitional patterns and inherent multimodal knowledge, ii) how to align both features into one unified LLM, minimize discrepancy while maximizing representation utility. To this end, we propose a multimodal LLM-enhanced framework TPAD, which extends a distillation paradigm to decouple and align transitional patterns for promoting MSBR. TPAD establishes parallel Knowledge-MLLM and Transfer-MLLM, where the former interprets item knowledge-reflected features and the latter extracts transition-aware features underneath sessions. A transitional pattern alignment module harnessing mutual information estimation theory unites two MLLMs, alleviating distribution discrepancy and distilling transitional patterns into modal representations. Extensive experiments on real-world datasets demonstrate the effectiveness of our framework.
arXiv:2504.11268v1 Announce Type: cross
Abstract: Model merging is a flexible and computationally tractable approach to merge single-task checkpoints into a multi-task model. Prior work has solely focused on constrained multi-task settings where there is a one-to-one mapping between a sample and a task, overlooking the paradigm where multiple tasks may operate on the same sample, e.g., scene understanding. In this paper, we focus on the multi-task setting with single-input-multiple-outputs (SIMO) and show that it qualitatively differs from the single-input-single-output model merging settings studied in the literature due to the existence of task-specific decoders and diverse loss objectives. We identify that existing model merging methods lead to significant performance degradation, primarily due to representation misalignment between the merged encoder and task-specific decoders. We propose two simple and efficient fixes for the SIMO setting to re-align the feature representation after merging. Compared to joint fine-tuning, our approach is computationally effective and flexible, and sheds light into identifying task relationships in an offline manner. Experiments on NYUv2, Cityscapes, and a subset of the Taskonomy dataset demonstrate: (1) task arithmetic suffices to enable multi-task capabilities; however, the representations generated by the merged encoder has to be re-aligned with the task-specific heads; (2) the proposed architecture rivals traditional multi-task learning in performance but requires fewer samples and training steps by leveraging the existence of task-specific models.
arXiv:2504.11406v1 Announce Type: cross
Abstract: The necessity of abundant annotated data and complex network architectures presents a significant challenge in deep-learning Salient Object Detection (deep SOD) and across the broader deep-learning landscape. This challenge is particularly acute in medical applications in developing countries with limited computational resources. Combining modern and classical techniques offers a path to maintaining competitive performance while enabling practical applications. Feature Learning from Image Markers (FLIM) methodology empowers experts to design convolutional encoders through user-drawn markers, with filters learned directly from these annotations. Recent findings demonstrate that coupling a FLIM encoder with an adaptive decoder creates a flyweight network suitable for SOD, requiring significantly fewer parameters than lightweight models and eliminating the need for backpropagation. Cellular Automata (CA) methods have proven successful in data-scarce scenarios but require proper initialization -- typically through user input, priors, or randomness. We propose a practical intersection of these approaches: using FLIM networks to initialize CA states with expert knowledge without requiring user interaction for each image. By decoding features from each level of a FLIM network, we can initialize multiple CAs simultaneously, creating a multi-level framework. Our method leverages the hierarchical knowledge encoded across different network layers, merging multiple saliency maps into a high-quality final output that functions as a CA ensemble. Benchmarks across two challenging medical datasets demonstrate the competitiveness of our multi-level CA approach compared to established models in the deep SOD literature.
arXiv:2504.10784v1 Announce Type: cross
Abstract: Autonomous systems deployed on edge devices face significant challenges, including resource constraints, real-time processing demands, and adapting to dynamic environments. This work introduces ATLASv2, a novel system that integrates a fine-tuned TinyLLM, real-time object detection, and efficient path planning to enable hierarchical, multi-task navigation and manipulation all on the edge device, Jetson Nano. ATLASv2 dynamically expands its navigable landmarks by detecting and localizing objects in the environment which are saved to its internal knowledge base to be used for future task execution. We evaluate ATLASv2 in real-world environments, including a handcrafted home and office setting constructed with diverse objects and landmarks. Results show that ATLASv2 effectively interprets natural language instructions, decomposes them into low-level actions, and executes tasks with high success rates. By leveraging generative AI in a fully on-board framework, ATLASv2 achieves optimized resource utilization with minimal prompting latency and power consumption, bridging the gap between simulated environments and real-world applications.
arXiv:2409.01007v3 Announce Type: replace
Abstract: This booklet, Unlocking the Wisdom of Multi-LLM Collaborative Intelligence, serves as an accessible introduction to the full volume The Path to Artificial General Intelligence. Through fourteen aphorisms, it distills the core principles of Multi-LLM Agent Collaborative Intelligence (MACI), a framework designed to coordinate multiple LLMs toward reasoning, planning, and decision-making that surpasses the capabilities of any single model. The booklet includes titles, abstracts, and introductions from each main chapter, along with the full content of the first two. The newly released third edition features significant enhancements to Chapters 6 through 9 and a revised preface responding to Yann LeCun's critique of AGI feasibility. While LeCun argues that LLMs lack grounding, memory, and planning, we propose that MACI's collaborative architecture, featuring multimodal agents in executive, legislative, and judicial roles, directly addresses these limitations. Chapters on SocraSynth, EVINCE, consciousness modeling, and behavior regulation demonstrate that reasoning systems grounded in structured interaction and checks and balances can produce more reliable, interpretable, and adaptive intelligence. By integrating complementary model strengths, including world modeling and multimodal perception, MACI enables a system-level intelligence that exceeds the sum of its parts. Like human institutions, progress in AI may depend less on isolated performance and more on coordinated judgment. Collaborative LLMs, not just larger ones, may chart the path toward artificial general intelligence.
arXiv:2504.10679v1 Announce Type: cross
Abstract: Brand reputation in the banking sector is maintained through insightful analysis of customer opinion on code-mixed and multilingual content. Conventional NLP models misclassify or ignore code-mixed text, when mix with low resource languages such as Sinhala-English and fail to capture domain-specific knowledge. This study introduces a hybrid NLP method to improve keyword extraction, content filtering, and aspect-based classification of banking content. Keyword extraction in English is performed with a hybrid approach comprising a fine-tuned SpaCy NER model, FinBERT-based KeyBERT embeddings, YAKE, and EmbedRank, which results in a combined accuracy of 91.2%. Code-mixed and Sinhala keywords are extracted using a fine-tuned XLM-RoBERTa model integrated with a domain-specific Sinhala financial vocabulary, and it results in an accuracy of 87.4%. To ensure data quality, irrelevant comment filtering was performed using several models, with the BERT-base-uncased model achieving 85.2% for English and XLM-RoBERTa 88.1% for Sinhala, which was better than GPT-4o, SVM, and keyword-based filtering. Aspect classification followed the same pattern, with the BERT-base-uncased model achieving 87.4% for English and XLM-RoBERTa 85.9% for Sinhala, both exceeding GPT-4 and keyword-based approaches. These findings confirm that fine-tuned transformer models outperform traditional methods in multilingual financial text analysis. The present framework offers an accurate and scalable solution for brand reputation monitoring in code-mixed and low-resource banking environments.
arXiv:2504.10636v1 Announce Type: cross
Abstract: We compare the performance of human and artificially intelligent (AI) decision makers in simple binary classification tasks where the optimal decision rule is given by Bayes Rule. We reanalyze choices of human subjects gathered from laboratory experiments conducted by El-Gamal and Grether and Holt and Smith. We confirm that while overall, Bayes Rule represents the single best model for predicting human choices, subjects are heterogeneous and a significant share of them make suboptimal choices that reflect judgement biases described by Kahneman and Tversky that include the ``representativeness heuristic'' (excessive weight on the evidence from the sample relative to the prior) and ``conservatism'' (excessive weight on the prior relative to the sample). We compare the performance of AI subjects gathered from recent versions of large language models (LLMs) including several versions of ChatGPT. These general-purpose generative AI chatbots are not specifically trained to do well in narrow decision making tasks, but are trained instead as ``language predictors'' using a large corpus of textual data from the web. We show that ChatGPT is also subject to biases that result in suboptimal decisions. However we document a rapid evolution in the performance of ChatGPT from sub-human performance for early versions (ChatGPT 3.5) to superhuman and nearly perfect Bayesian classifications in the latest versions (ChatGPT 4o).
arXiv:2504.11343v1 Announce Type: cross
Abstract: Reinforcement learning (RL) has become a prevailing approach for fine-tuning large language models (LLMs) on complex reasoning tasks. Among recent methods, GRPO stands out for its empirical success in training models such as DeepSeek-R1, yet the sources of its effectiveness remain poorly understood. In this work, we revisit GRPO from a reinforce-like algorithm perspective and analyze its core components. Surprisingly, we find that a simple rejection sampling baseline, RAFT, which trains only on positively rewarded samples, yields competitive performance than GRPO and PPO. Our ablation studies reveal that GRPO's main advantage arises from discarding prompts with entirely incorrect responses, rather than from its reward normalization. Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples. Reinforce-Rej improves KL efficiency and stability, serving as a lightweight yet effective alternative to more complex RL algorithms. We advocate RAFT as a robust and interpretable baseline, and suggest that future advances should focus on more principled designs for incorporating negative samples, rather than relying on them indiscriminately. Our findings provide guidance for future work in reward-based LLM post-training.
arXiv:2504.10797v1 Announce Type: cross
Abstract: Across cultures, names tell a lot about their bearers as they carry deep personal and cultural significance. Names also serve as powerful signals of gender, race, and status in the social hierarchy - a pecking order in which individual positions shape others' expectations on their perceived competence and worth. With the widespread adoption of LLMs and as names are often an input for LLMs, it is crucial to evaluate whether LLMs may sort people into status positions based on first and last names and, if so, whether it is in an unfair, biased fashion. While prior work has primarily investigated biases in first names, little attention has been paid to last names and even less to the combined effects of first and last names. In this study, we conduct a large-scale analysis of name variations across 5 ethnicities to examine how AI exhibits name biases. Our study investigates three key characteristics of inequality and finds that LLMs reflect and reinforce status hierarchies based on names that signal gender and ethnicity as they encode differential expectations of competence, leadership, and economic potential. Contrary to the common assumption that AI tends to favor Whites, we show that East and, in some contexts, South Asian names receive higher rankings. We also disaggregate Asians, a population projected to be the largest immigrant group in the U.S. by 2055. Our results challenge the monolithic Asian model minority assumption, illustrating a more complex and stratified model of bias. Gender moderates biases, with girls facing unfair disadvantages in certain racial groups. Additionally, spanning cultural categories by adopting Western first names improves AI-perceived status for East and Southeast Asian students, particularly for girls. Our findings underscore the importance of intersectional and more nuanced understandings of race, gender, and mixed identities in the evaluation of LLMs.
arXiv:2504.10888v1 Announce Type: cross
Abstract: Adversarial patches are widely used to evaluate the robustness of object detection systems in real-world scenarios. These patches were initially designed to deceive single-modal detectors (e.g., visible or infrared) and have recently been extended to target visible-infrared dual-modal detectors. However, existing dual-modal adversarial patch attacks have limited attack effectiveness across diverse physical scenarios. To address this, we propose CDUPatch, a universal cross-modal patch attack against visible-infrared object detectors across scales, views, and scenarios. Specifically, we observe that color variations lead to different levels of thermal absorption, resulting in temperature differences in infrared imaging. Leveraging this property, we propose an RGB-to-infrared adapter that maps RGB patches to infrared patches, enabling unified optimization of cross-modal patches. By learning an optimal color distribution on the adversarial patch, we can manipulate its thermal response and generate an adversarial infrared texture. Additionally, we introduce a multi-scale clipping strategy and construct a new visible-infrared dataset, MSDrone, which contains aerial vehicle images in varying scales and perspectives. These data augmentation strategies enhance the robustness of our patch in real-world conditions. Experiments on four benchmark datasets (e.g., DroneVehicle, LLVIP, VisDrone, MSDrone) show that our method outperforms existing patch attacks in the digital domain. Extensive physical tests further confirm strong transferability across scales, views, and scenarios.
arXiv:2504.10500v1 Announce Type: cross
Abstract: This paper introduces a cutting-edge method for enhancing recommender systems through the integration of generative self-supervised learning (SSL) with a Residual Graph Transformer. Our approach emphasizes the importance of superior data enhancement through the use of pertinent pretext tasks, automated through rationale-aware SSL to distill clear ways of how users and items interact. The Residual Graph Transformer incorporates a topology-aware transformer for global context and employs residual connections to improve graph representation learning. Additionally, an auto-distillation process refines self-supervised signals to uncover consistent collaborative rationales. Experimental evaluations on multiple datasets demonstrate that our approach consistently outperforms baseline methods.
arXiv:2503.09309v2 Announce Type: replace-cross
Abstract: Incentive design is a popular framework for guiding agents' learning dynamics towards desired outcomes by providing additional payments beyond intrinsic rewards. However, most existing works focus on a finite, small set of agents or assume complete knowledge of the game, limiting their applicability to real-world scenarios involving large populations and model uncertainty. To address this gap, we study the design of steering rewards in Mean-Field Games (MFGs) with density-independent transitions, where both the transition dynamics and intrinsic reward functions are unknown. This setting presents non-trivial challenges, as the mediator must incentivize the agents to explore for its model learning under uncertainty, while simultaneously steer them to converge to desired behaviors without incurring excessive incentive payments. Assuming agents exhibit no(-adaptive) regret behaviors, we contribute novel optimistic exploration algorithms. Theoretically, we establish sub-linear regret guarantees for the cumulative gaps between the agents' behaviors and the desired ones. In terms of the steering cost, we demonstrate that our total incentive payments incur only sub-linear excess, competing with a baseline steering strategy that stabilizes the target policy as an equilibrium. Our work presents an effective framework for steering agents behaviors in large-population systems under uncertainty.
arXiv:2504.10738v1 Announce Type: cross
Abstract: The rapid growth of intelligent connected vehicles (ICVs) and integrated vehicle-road-cloud systems has increased the demand for accurate, real-time HD map updates. However, ensuring map reliability remains challenging due to inconsistencies in crowdsourced data, which suffer from motion blur, lighting variations, adverse weather, and lane marking degradation. This paper introduces CleanMAP, a Multimodal Large Language Model (MLLM)-based distillation framework designed to filter and refine crowdsourced data for high-confidence HD map updates. CleanMAP leverages an MLLM-driven lane visibility scoring model that systematically quantifies key visual parameters, assigning confidence scores (0-10) based on their impact on lane detection. A novel dynamic piecewise confidence-scoring function adapts scores based on lane visibility, ensuring strong alignment with human evaluations while effectively filtering unreliable data. To further optimize map accuracy, a confidence-driven local map fusion strategy ranks and selects the top-k highest-scoring local maps within an optimal confidence range (best score minus 10%), striking a balance between data quality and quantity. Experimental evaluations on a real-world autonomous vehicle dataset validate CleanMAP's effectiveness, demonstrating that fusing the top three local maps achieves the lowest mean map update error of 0.28m, outperforming the baseline (0.37m) and meeting stringent accuracy thresholds (
arXiv:2504.11159v1 Announce Type: new
Abstract: Time series are ubiquitous in domains such as energy forecasting, healthcare, and industry. Using AI systems, some tasks within these domains can be efficiently handled. Explainable AI (XAI) aims to increase the reliability of AI solutions by explaining model reasoning. For time series, many XAI methods provide point- or sequence-based attribution maps. These methods explain model reasoning in terms of low-level patterns. However, they do not capture high-level patterns that may also influence model reasoning. We propose a concept-based method to provide explanations in terms of these high-level patterns. In this paper, we present C-SHAP for time series, an approach which determines the contribution of concepts to a model outcome. We provide a general definition of C-SHAP and present an example implementation using time series decomposition. Additionally, we demonstrate the effectiveness of the methodology through a use case from the energy domain.
arXiv:2411.10329v2 Announce Type: replace-cross
Abstract: In recent years, text-to-image (T2I) generation models have made significant progress in generating high-quality images that align with text descriptions. However, these models also face the risk of unsafe generation, potentially producing harmful content that violates usage policies, such as explicit material. Existing safe generation methods typically focus on suppressing inappropriate content by erasing undesired concepts from visual representations, while neglecting to sanitize the textual representation. Although these methods help mitigate the risk of misuse to some extent, their robustness remains insufficient when dealing with adversarial attacks.
Given that semantic consistency between input text and output image is a core requirement of T2I models, we identify that textual representations are likely the primary source of unsafe generation. To this end, we propose Embedding Sanitizer (ES), which enhances the safety of T2I models by sanitizing inappropriate concepts in prompt embeddings. To our knowledge, ES is the first interpretable safe generation framework that assigns a score to each token in the prompt to indicate its potential harmfulness. In addition, ES adopts a plug-and-play modular design, offering compatibility for seamless integration with various T2I models and other safeguards. Evaluations on five prompt benchmarks show that ES outperforms eleven existing safeguard baselines, achieving state-of-the-art robustness while maintaining high-quality image generation.
arXiv:2504.10540v1 Announce Type: cross
Abstract: Diffusion models have demonstrated remarkable success in generative tasks, yet their iterative denoising process results in slow inference, limiting their practicality. While existing acceleration methods exploit the well-known U-shaped similarity pattern between adjacent steps through caching mechanisms, they lack theoretical foundation and rely on simplistic computation reuse, often leading to performance degradation. In this work, we provide a theoretical understanding by analyzing the denoising process through the second-order Adams-Bashforth method, revealing a linear relationship between the outputs of consecutive steps. This analysis explains why the outputs of adjacent steps exhibit a U-shaped pattern. Furthermore, extending Adams-Bashforth method to higher order, we propose a novel caching-based acceleration approach for diffusion models, instead of directly reusing cached results, with a truncation error bound of only \(O(h^k)\) where $h$ is the step size. Extensive validation across diverse image and video diffusion models (including HunyuanVideo and FLUX.1-dev) with various schedulers demonstrates our method's effectiveness in achieving nearly $3\times$ speedup while maintaining original performance levels, offering a practical real-time solution without compromising generation quality.
arXiv:2504.10903v1 Announce Type: cross
Abstract: Reasoning models have demonstrated remarkable progress in solving complex and logic-intensive tasks by generating extended Chain-of-Thoughts (CoTs) prior to arriving at a final answer. Yet, the emergence of this "slow-thinking" paradigm, with numerous tokens generated in sequence, inevitably introduces substantial computational overhead. To this end, it highlights an urgent need for effective acceleration. This survey aims to provide a comprehensive overview of recent advances in efficient reasoning. It categorizes existing works into three key directions: (1) shorter - compressing lengthy CoTs into concise yet effective reasoning chains; (2) smaller - developing compact language models with strong reasoning capabilities through techniques such as knowledge distillation, other model compression techniques, and reinforcement learning; and (3) faster - designing efficient decoding strategies to accelerate inference. A curated collection of papers discussed in this survey is available in our GitHub repository.
arXiv:2504.03624v3 Announce Type: replace-cross
Abstract: As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.
arXiv:2504.04222v2 Announce Type: replace-cross
Abstract: Machine learning (ML) powered network traffic analysis has been widely used for the purpose of threat detection. Unfortunately, their generalization across different tasks and unseen data is very limited. Large language models (LLMs), known for their strong generalization capabilities, have shown promising performance in various domains. However, their application to the traffic analysis domain is limited due to significantly different characteristics of network traffic. To address the issue, in this paper, we propose TrafficLLM, which introduces a dual-stage fine-tuning framework to learn generic traffic representation from heterogeneous raw traffic data. The framework uses traffic-domain tokenization, dual-stage tuning pipeline, and extensible adaptation to help LLM release generalization ability on dynamic traffic analysis tasks, such that it enables traffic detection and traffic generation across a wide range of downstream tasks. We evaluate TrafficLLM across 10 distinct scenarios and 229 types of traffic. TrafficLLM achieves F1-scores of 0.9875 and 0.9483, with up to 80.12% and 33.92% better performance than existing detection and generation methods. It also shows strong generalization on unseen traffic with an 18.6% performance improvement. We further evaluate TrafficLLM in real-world scenarios. The results confirm that TrafficLLM is easy to scale and achieves accurate detection performance on enterprise traffic.
arXiv:2504.11369v1 Announce Type: cross
Abstract: Open Large Language Models (OLLMs) are increasingly leveraged in generative AI applications, posing new challenges for detecting their outputs. We propose OpenTuringBench, a new benchmark based on OLLMs, designed to train and evaluate machine-generated text detectors on the Turing Test and Authorship Attribution problems. OpenTuringBench focuses on a representative set of OLLMs, and features a number of challenging evaluation tasks, including human/machine-manipulated texts, out-of-domain texts, and texts from previously unseen models. We also provide OTBDetector, a contrastive learning framework to detect and attribute OLLM-based machine-generated texts. Results highlight the relevance and varying degrees of difficulty of the OpenTuringBench tasks, with our detector achieving remarkable capabilities across the various tasks and outperforming most existing detectors. Resources are available on the OpenTuringBench Hugging Face repository at https://huggingface.co/datasets/MLNTeam-Unical/OpenTuringBench
arXiv:2504.11045v1 Announce Type: cross
Abstract: As autonomous systems become increasingly prevalent in daily life, ensuring their safety is paramount. Control Barrier Functions (CBFs) have emerged as an effective tool for guaranteeing safety; however, manually designing them for specific applications remains a significant challenge. With the advent of deep learning techniques, recent research has explored synthesizing CBFs using neural networks-commonly referred to as neural CBFs. This paper introduces a novel class of neural CBFs that leverages a physics-inspired neural network framework by incorporating Zubov's Partial Differential Equation (PDE) within the context of safety. This approach provides a scalable methodology for synthesizing neural CBFs applicable to high-dimensional systems. Furthermore, by utilizing reciprocal CBFs instead of zeroing CBFs, the proposed framework allows for the specification of flexible, user-defined safe regions. To validate the effectiveness of the approach, we present case studies on three different systems: an inverted pendulum, autonomous ground navigation, and aerial navigation in obstacle-laden environments.
arXiv:2504.10478v2 Announce Type: replace-cross
Abstract: We investigate a failure mode that arises during the training of reasoning models, where the diversity of generations begins to collapse, leading to suboptimal test-time scaling. Notably, the Pass@1 rate reliably improves during supervised finetuning (SFT), but Pass@k rapidly deteriorates. Surprisingly, a simple intervention of interpolating the weights of the latest SFT checkpoint with an early checkpoint, otherwise known as WiSE-FT, almost completely recovers Pass@k while also improving Pass@1. The WiSE-FT variant achieves better test-time scaling (Best@k, majority vote) and achieves superior results with less data when tuned further by reinforcement learning. Finally, we find that WiSE-FT provides complementary performance gains that cannot be achieved only through diversity-inducing decoding strategies, like temperature scaling. We formalize a bias-variance tradeoff of Pass@k with respect to the expectation and variance of Pass@1 over the test distribution. We find that WiSE-FT can reduce bias and variance simultaneously, while temperature scaling inherently trades-off between bias and variance.
arXiv:2407.19631v3 Announce Type: replace
Abstract: How can intelligent machines assess their competency to complete a task? This question has come into focus for autonomous systems that algorithmically make decisions under uncertainty. We argue that machine self-confidence -- a form of meta-reasoning based on self-assessments of system knowledge about the state of the world, itself, and ability to reason about and execute tasks -- leads to many computable and useful competency indicators for such agents. This paper presents our body of work, so far, on this concept in the form of the Factorized Machine Self-confidence (FaMSeC) framework, which holistically considers several major factors driving competency in algorithmic decision-making: outcome assessment, solver quality, model quality, alignment quality, and past experience. In FaMSeC, self-confidence indicators are derived via 'problem-solving statistics' embedded in Markov decision process solvers and related approaches. These statistics come from evaluating probabilistic exceedance margins in relation to certain outcomes and associated competency standards specified by an evaluator. Once designed, and evaluated, the statistics can be easily incorporated into autonomous agents and serve as indicators of competency. We include detailed descriptions and examples for Markov decision process agents, and show how outcome assessment and solver quality factors can be found for a range of tasking contexts through novel use of meta-utility functions, behavior simulations, and surrogate prediction models. Numerical evaluations are performed to demonstrate that FaMSeC indicators perform as desired (references to human subject studies beyond the scope of this paper are provided).
arXiv:2504.10823v1 Announce Type: cross
Abstract: Navigating high-stakes dilemmas involving conflicting values is challenging even for humans, let alone for AI. Yet prior work in evaluating the reasoning capabilities of large language models (LLMs) in such situations has been limited to everyday scenarios. To close this gap, this work first introduces CLASH (Character perspective-based LLM Assessments in Situations with High-stakes), a meticulously curated dataset consisting of 345 high-impact dilemmas along with 3,795 individual perspectives of diverse values. In particular, we design CLASH in a way to support the study of critical aspects of value-based decision-making processes which are missing from prior work, including understanding decision ambivalence and psychological discomfort as well as capturing the temporal shifts of values in characters' perspectives. By benchmarking 10 open and closed frontier models, we uncover several key findings. (1) Even the strongest models, such as GPT-4o and Claude-Sonnet, achieve less than 50% accuracy in identifying situations where the decision should be ambivalent, while they perform significantly better in clear-cut scenarios. (2) While LLMs reasonably predict psychological discomfort as marked by human, they inadequately comprehend perspectives involving value shifts, indicating a need for LLMs to reason over complex values. (3) Our experiments also reveal a significant correlation between LLMs' value preferences and their steerability towards a given value. (4) Finally, LLMs exhibit greater steerability when engaged in value reasoning from a third-party perspective, compared to a first-person setup, though certain value pairs benefit uniquely from the first-person framing.
arXiv:2504.11423v1 Announce Type: cross
Abstract: Diffusion models have achieved outstanding image generation by reversing a forward noising process to approximate true data distributions. During training, these models predict diffusion scores from noised versions of true samples in a single forward pass, while inference requires iterative denoising starting from white noise. This training-inference divergences hinder the alignment between inference and training data distributions, due to potential prediction biases and cumulative error accumulation. To address this problem, we propose an intuitive but effective fine-tuning framework, called Adversarial Diffusion Tuning (ADT), by stimulating the inference process during optimization and aligning the final outputs with training data by adversarial supervision. Specifically, to achieve robust adversarial training, ADT features a siamese-network discriminator with a fixed pre-trained backbone and lightweight trainable parameters, incorporates an image-to-image sampling strategy to smooth discriminative difficulties, and preserves the original diffusion loss to prevent discriminator hacking. In addition, we carefully constrain the backward-flowing path for back-propagating gradients along the inference path without incurring memory overload or gradient explosion. Finally, extensive experiments on Stable Diffusion models (v1.5, XL, and v3), demonstrate that ADT significantly improves both distribution alignment and image quality.