论文检索

Deep Learning · Large Language Models

Qiushi Huang, Tom Ko, Lilian Tang, Yu Zhang

We propose a Competitive Low-Rank Adaptation (ComLoRA) framework to address the limitations of the LoRA method, which either lacks capacity with a single rank-$r$ LoRA or risks inefficiency and overfitting with a larger rank-$Kr$ LoRA, where $K$ is an integer larger than 1. The proposed ComLoRA method initializes $K$ distinct LoRA components, each with rank $r$, and allows them to compete during training. This competition drives each LoRA component to outperform the others, improving overall model performance. The best-performing LoRA is selected based on validation metrics, ensuring that the final model outperforms a single rank-$r$ LoRA and matches the effectiveness of a larger rank-$Kr$ LoRA, all while avoiding extra computational overhead during inference. To the best of our knowledge, this is the first work to introduce and explore competitive learning in the context of LoRA optimization. The ComLoRA's code is available at https://github.com/hqsiswiliam/comlora.

Miscellaneous Aspects of Machine Learning · Supervised Learning

Gathika Ratnayaka, James Nichols, Qing Wang

Partial graph matching extends traditional graph matching by allowing some nodes to remain unmatched, enabling applications in more complex scenarios. However, this flexibility introduces additional complexity, as both the subset of nodes to match and the optimal mapping must be determined. While recent studies have explored deep learning techniques for partial graph matching, a significant limitation remains: the absence of an optimization objective that fully captures the problem’s intrinsic nature while enabling efficient solutions. In this paper, we propose a novel optimization framework for partial graph matching, inspired by optimal partial transport. Our approach formulates an objective that enables partial assignments while incorporating matching biases, using weighted total variation as the divergence function to guarantee optimal partial assignments. Our method can achieve efficient, exact solutions within cubic worst case time complexity. Our contributions are threefold: (i) we introduce a novel optimization objective that balances matched and unmatched nodes; (ii) we establish a connection between partial graph matching and linear sum assignment problem, enabling efficient solutions; (iii) we propose a deep graph matching architecture with a novel partial matching loss, providing an end-to-end solution. The empirical evaluations on standard graph matching benchmarks demonstrate the efficacy of the proposed approach.

Miscellaneous Aspects of Machine Learning · General Machine Learning Techniques

Zihan Yu, Jingtao Ding, Yong Li, Depeng Jin

Symbolic regression, a task discovering the formula best fitting the given data, is typically based on the heuristical search. These methods usually update candidate formulas to obtain new ones with lower prediction errors iteratively. However, since formulas with similar function shapes may have completely different symbolic forms, the prediction error does not decrease monotonously as the search approaches the target formula, causing the low recovery rate of existing methods. To solve this problem, we propose a novel search objective based on the minimum description length, which reflects the distance from the target and decreases monotonically as the search approaches the correct form of the target formula. To estimate the minimum description length of any input data, we design a neural network, MDLformer, which enables robust and scalable estimation through large-scale training. With the MDLformer's output as the search objective, we implement a symbolic regression method, SR4MDL, that can effectively recover the correct mathematical form of the formula. Extensive experiments illustrate its excellent performance in recovering formulas from data. Our method successfully recovers around 50 formulas across two benchmark datasets comprising 133 problems, outperforming state-of-the-art methods by 43.92%. Experiments on 122 unseen black-box problems further demonstrate its generalization performance. We release our code at https://github.com/tsinghua-fib-lab/SR4MDL .

Deep Learning · Large Language Models

Sohan Patnaik, Milan Aggarwal, Sumit Bhatia, Balaji Krishnamurthy

Very large language models (LLMs) such as GPT-4 have shown the ability to handle complex tasks by generating and self-refining step-by-step rationales. Smaller language models (SLMs), typically with < 13B parameters, have been improved by using the data generated from very-large LMs through knowledge distillation. However, various practical constraints such as API costs, copyright, legal and ethical policies restrict using large (often opaque) models to train smaller models for commercial use. Limited success has been achieved at improving the ability of an SLM to explore the space of possible rationales and evaluate them by itself through self-deliberation. To address this, we propose COALITION, a trainable framework that facilitates interaction between two variants of the same SLM and trains them to generate and refine rationales optimized for the end-task. The variants exhibit different behaviors to produce a set of diverse candidate rationales during the generation and refinement steps. The model is then trained via Selective Rationale Optimization (SRO) to prefer generating rationale candidates that maximize the likelihood of producing the ground-truth answer. During inference, COALITION employs a controller to select the suitable variant for generating and refining the rationales. On five different datasets covering mathematical problems, commonsense reasoning, and natural language inference, COALITION outperforms several baselines by up to 5%. Our ablation studies reveal that cross-communication between the two variants performs better than using the single model to self-refine the rationales. We also demonstrate the applicability of COALITION for LMs of varying scales (4B to 14B parameters) and model families (Mistral, Llama, Qwen, Phi). We release the code for this work here.

Applications · Computer Vision

Isabella Liu, Hao Su, Xiaolong Wang

Modern 3D engines and graphics pipelines require mesh as a memory-efficient representation, which allows efficient rendering, geometry processing, texture editing, and many other downstream operations. However, it is still highly difficult to obtain high-quality mesh in terms of detailed structure and time consistency from dynamic observations. To this end, we introduce Dynamic Gaussians Mesh (DG-Mesh), a framework to reconstruct a high-fidelity and time-consistent mesh from dynamic input. Our work leverages the recent advancement in 3D Gaussian Splatting to construct the mesh sequence with temporal consistency from dynamic observations. Building on top of this representation, DG-Mesh recovers high-quality meshes from the Gaussian points and can track the mesh vertices over time, which enables applications such as texture editing on dynamic objects. We introduce the Gaussian-Mesh Anchoring, which encourages evenly distributed Gaussians, resulting better mesh reconstruction through mesh-guided densification and pruning on the deformed Gaussians. By applying cycle-consistent deformation between the canonical and the deformed space, we can project the anchored Gaussian back to the canonical space and optimize Gaussians across all time frames. During the evaluation on different datasets, DG-Mesh provides significantly better mesh reconstruction and rendering than baselines.

Social Aspects · Everything Else

Javier Rando

Jailbreaks are becoming a new ImageNet competition instead of helping us better understand LLM security. This blogpost surveys the jailbreak literature to extract the most important contributions and encourages the community to revisit their choices and focus on research that can uncover new security vulnerabilities.

Deep Learning · Large Language Models

Yuze Zhao, Tianyun Ji, Wenjun Feng, Zhenya Huang, Qi Liu, Zhiding Liu, Yixiao Ma, Kai Zhang, Enhong Chen

The reasoning abilities are one of the most enigmatic and captivating aspects of large language models (LLMs). Numerous studies are dedicated to exploring and expanding the boundaries of this reasoning capability. However, tasks that embody both reasoning and recall characteristics are often overlooked. In this paper, we introduce such a novel task, code reasoning, to provide a new perspective for the reasoning abilities of LLMs.We summarize three meta-benchmarks based on established forms of logical reasoning, and instantiate these into eight specific benchmark tasks. Our testing on these benchmarks reveals that LLMs continue to struggle with identifying satisfactory reasoning pathways. Additionally, we present a new pathway exploration pipeline inspired by human intricate problem-solving methods. This Reflective Hypothesis Decomposition and Amendment (RHDA) pipeline consists of the following iterative steps: (1) Proposing potential hypotheses based on observations and decomposing them; (2) Utilizing tools to validate hypotheses and reflection outcomes; (3) Revising hypothesis in light of observations. Our approach effectively mitigates logical chain collapses arising from forgetting or hallucination issues in multi-step reasoning, resulting in performance gains of up to $3\times$. Finally, we expanded this pipeline by applying it to simulate complex household tasks in real-world scenarios, specifically in VirtualHome, enhancing the handling of failure cases. We release our code and all of results at https://github.com/TnTWoW/code_reasoning.

Applications · Neuroscience, Cognitive Science

Haixin Zhong, Haoyu Wang, Wei Dai, Yuchao Huang, Mingyi Huang, Rubin Wang, Anna Roe, Yuguo Yu

Extensive experimental studies have shown that in lower mammals, neuronal orientation preference in the primary visual cortex is organized in disordered "salt-and-pepper" organizations. In contrast, higher-order mammals display a continuous variation in orientation preference, forming pinwheel-like structures. Despite these observations, the spiking mechanisms underlying the emergence of these distinct topological structures and their functional roles in visual processing remain poorly understood. To address this, we developed a self-evolving spiking neural network model with Hebbian plasticity, trained using physiological parameters characteristic of rodents, cats, and primates, including retinotopy, neuronal morphology, and connectivity patterns. Our results identify critical factors, such as the degree of input visual field overlap, neuronal connection range, and the balance between localized connectivity and long-range competition, that determine the emergence of either salt-and-pepper or pinwheel-like topologies. Furthermore, we demonstrate that pinwheel structures exhibit lower wiring costs and enhanced sparse coding capabilities compared to salt-and-pepper organizations. They also maintain greater coding robustness against noise in naturalistic visual stimuli. These findings suggest that such topological structures confer significant computational advantages in visual processing and highlight their potential application in the design of brain-inspired deep learning networks and algorithms.

Deep Learning · Graph Neural Networks

Bin Liang, Shiwei Chen, Lin Gui, HUI WANG, Yue Yu, Ruifeng Xu, Kam-Fai Wong

Self-supervised learning (SSL) has shown great potential in learning generalizable representations for graph-structured data. However, existing SSL-based graph pre-training methods largely focus on improving graph representations by learning the structure information based on disturbing or reconstructing graphs, which ignores an important issue: the importance of different nodes in the graph structure may vary. To fill this gap, we propose a Centrality-guided Graph Pre-training (CenPre) framework to integrate the distinct importance of nodes in graph structure into the corresponding representations of nodes based on the centrality in graph theory. In this way, the different roles played by different nodes can be effectively leveraged when learning graph structure. The proposed CenPre contains three modules for node representation pre-training and alignment. The node-level importance learning module fuses the fine-grained node importance into node representation based on degree centrality, allowing the aggregation of node representations with equal/similar importance. The graph-level importance learning module characterizes the importance between all nodes in the graph based on eigenvector centrality, enabling the exploitation of graph-level structure similarities/differences when learning node representation. Finally, a representation alignment module aligns the pre-trained node representation using the original one, essentially allowing graph representations to learn structural information without losing their original semantic information, thereby leading to better graph representations. Extensive experiments on a series of real-world datasets demonstrate that the proposed CenPre outperforms the state-of-the-art baselines in the tasks of node classification, link prediction, and graph classification.

Hong Qian, Yiyi Zhu, Xiang Shu, Shuo Liu, Yaolin Wen, Xin An, Huakang LU, Aimin Zhou, Ke Tang, Yang Yu

Black-box optimization aims to find the optima through building a model close to the black-box objective function based on function value evaluation. However, in many real-world tasks, such as the design of molecular formulas and mechanical structures, it is perilous, costly, or even infeasible to evaluate the objective function value of an actively sampled solution. In this situation, optimization can only be conducted via utilizing offline historical data, which yields offline black-box optimization. Different from the traditional goal that is to pursue the optimal solution, this paper emphasizes that the goal of offline optimization is to stably surpass the offline dataset during optimization procedure. Although benchmarks called Design-Bench already exist in this emerging field, it can hardly evaluate the stability of offline optimization and mainly provides real-world offline tasks and the corresponding offline datasets. To this end, this paper proposes benchmarks named SOO-Bench (i.e., Stable Offline Optimization Benchmarks) for offline black-box optimization algorithms, so as to systematically evaluate the stability of surpassing the offline dataset under different data distributions. Along with SOO-Bench, we also propose a stability indicator to measure the degree of stability. Specifically, SOO-Bench includes various real-world offline optimization tasks and offline datasets under different data distributions, involving the fields of satellites, materials science, structural mechanics, and automobile manufacturing. Empirically, baseline and state-of-the-art algorithms are tested and analyzed on SOO-Bench. Hopefully, SOO-Bench is expected to serve as a catalyst for the rapid developments of more novel and stable offline optimization methods. The code is available at \url{https://github.com/zhuyiyi-123/SOO-Bench}.

Dmitry Kamzolov, Artem Agafonov, Dmitry Pasechnyuk, Alexander Gasnikov, Martin Takáč

Second-order methods for convex optimization outperform first-order methods in terms of theoretical iteration convergence, achieving rates up to $O(k^{-5})$ for highly-smooth functions. However, their practical performance and applications are limited due to their multi-level structure and implementation complexity. In this paper, we present new results on high-order optimization methods, supported by their practical performance. First, we show that the basic high-order methods, such as the Cubic Regularized Newton Method, exhibit global superlinear convergence for $\mu$-strongly star-convex functions, a class that includes $\mu$-strongly convex functions and some non-convex functions. Theoretical convergence results are both inspired and supported by the practical performance of these methods. Secondly, we propose a practical version of the Nesterov Accelerated Tensor method, called NATA. It significantly outperforms the classical variant and other high-order acceleration techniques in practice. The convergence of NATA is also supported by theoretical results. Finally, we introduce an open-source computational library for high-order methods, called OPTAMI. This library includes various methods, acceleration techniques, and subproblem solvers, all implemented as PyTorch optimizers, thereby facilitating the practical application of high-order methods to a wide range of optimization problems. We hope this library will simplify research and practical comparison of methods beyond first-order.

caigao jiang, Xiang Shu, Hong Qian, Xingyu Lu, JUN ZHOU, Aimin Zhou, Yang Yu

Optimization problems are prevalent across various scenarios. Formulating and then solving optimization problems described by natural language often requires highly specialized human expertise, which could block the widespread application of optimization-based decision making. To automate problem formulation and solving, leveraging large language models (LLMs) has emerged as a potential way. However, this kind of approach suffers from the issue of optimization generalization. Namely, the accuracy of most current LLM-based methods and the generality of optimization problem types that they can model are still limited. In this paper, we propose a unified learning-based framework called LLMOPT to boost optimization generalization. Starting from the natural language descriptions of optimization problems and a pre-trained LLM, LLMOPT constructs the introduced five-element formulation as a universal model for learning to define diverse optimization problem types. Then, LLMOPT employs the multi-instruction tuning to enhance both problem formalization and solver code generation accuracy and generality. After that, to prevent hallucinations in LLMs, such as sacrificing solving accuracy to avoid execution errors, the model alignment and self-correction mechanism are adopted in LLMOPT. We evaluate the optimization generalization ability of LLMOPT and compared methods across six real-world datasets covering roughly 20 fields such as health, environment, energy and manufacturing, etc. Extensive experiment results show that LLMOPT is able to model various optimization problem types such as linear/nonlinear programming, mixed integer programming, and combinatorial optimization, and achieves a notable 11.08% average solving accuracy improvement compared with the state-of-the-art methods. The code is available at https://github.com/caigaojiang/LLMOPT.

Letian Peng, Chenyang An, Jingbo Shang

Language model (LM) decoding is based on the next-token prediction (NTP) probability distribution. For neural LMs (e.g., Transformer-based), NTP distribution isessentially a softmax-regularized dot product between an encoded input context(query) and fixed vocabulary representations (keys). In this paper, we study theeffect of the key distribution on the NTP distribution, with a focus on whetherthe similarity between keys will trigger spurious correlations in NTP. Throughknowledge-probing tasks, we show that in the NTP distribution, the few top-rankedtokens are typically accurate. However, the middle-ranked prediction is highly biasedtowards the tokens that are distributionally (not necessarily semantically) similar tothese top ones. For instance, if “P” is predicted as the top-1 token, “A”-“Z” will allbe ranked high in NTP, no matter whether they can lead to correct decoding results.This hurts the sampling diversity and makes the sampling of correct, long-tailresults hopeless and noisy. We attempt to alleviate this issue via a novel in-contextmethod that iteratively pushes the query representation away from explored regions.Specifically, we include the explored decoding results in the context and promptthe LM to generate something else, which encourages the LM to produce a queryrepresentation that has small dot products with explored keys. Experiments onknowledge-probing tasks show that our method leads to efficient navigation awayfrom explored keys to correct new keys. We further extend our method to open-ended and chain-of-thought (for reasoning) generation. Experiment results showthat ICN contributes to better generation diversity and improved self-consistencyvoting performance. Finally, we discuss potential training issues caused by thefixed key space together with the challenges and possible ways to address them infuture research.

Social Aspects · Accountability, Transparency and Interpretability

Chase Walker, Sumit Jha, Rickard Ewetz

Attribution algorithms explain computer vision models by attributing the model response to pixels within the input. Existing attribution methods generate explanations by combining transformations of internal model representations such as class activation maps, gradients, attention, or relevance scores. The effectiveness of an attribution map is measured using attribution quality metrics. This leads us to pose the following question: if attribution methods are assessed using attribution quality metrics, why are the metrics not used to generate the attributions? In response to this question, we propose a Metric-Driven Attribution for explaining Vision Transformers (ViT) called MDA. Guided by attribution quality metrics, the method creates attribution maps by performing patch order and patch magnitude optimization across all patch tokens. The first step orders the patches in terms of importance and the second step assigns the magnitude to each patch while preserving the patch order. Moreover, MDA can provide a smooth trade-off between sparse and dense attributions by modifying the optimization objective. Experimental evaluation demonstrates the proposed MDA method outperforms $7$ existing ViT attribution methods by an average of $12\%$ across $12$ attribution metrics on the ImageNet dataset for the ViT-base $16 \times 16$, ViT-tiny $16 \times 16$, and ViT-base $32 \times 32$ models. Code is publicly available at https://github.com/chasewalker26/MDA-Metric-Driven-Attributions-for-ViT.

Reinforcement Learning · Everything Else

Gust Verbruggen, Ashish Tiwari, Mukul Singh, Vu Le, Sumit Gulwani

Large language models (LLMs) can generate code from examples without being limited to a DSL, but they lack search, as sampled programs are independent.In this paper, we use an LLM as a policy that generates lines of code and then join these lines of code to let the LLM implicitly estimate the value of each of these lines in its next iteration.We further guide the policy and value estimation by executing each line and annotating it with its results on the given examples. This allows us to search for programs within a single (expanding) prompt until a sound program is found, by letting the policy reason in both the syntactic (code) and semantic (execution) space.We evaluate within-prompt search on straight-line Python code generation using five benchmarks across different domains (strings, lists, and arbitrary Python programming problems).We show that the model uses the execution results to guide the search and that within-prompt search performs well at low token budgets.We also analyze how the model behaves as a policy and value, show that it can parallelize the search, and that it can implicitly backtrack over earlier generations.

Divyam Madaan, Sumit Chopra, Kyunghyun Cho

Advancements in language models has spurred an increasing interest in multi-modal AI — models that process and understand information across multiple forms of data, such as text, images and audio. While the goal is to emulate human-like ability to handle diverse information, a key question is: do human-defined modalities align with machine perception? If not, how does this misalignment affect AI performance? In this blog, we examine these questions by reflecting on the progress made by the community in developing multi-modal benchmarks and architectures, highlighting their limitations. By reevaluating our definitions and assumptions, we propose ways to better handle multi-modal data by building models that analyze and combine modality contributions both independently and jointly with other modalities.

Deep Learning · Large Language Models

Xu-Hui Liu, Yali Du, Jun Wang, Yang Yu

Training Large Language Models (LLMs) poses significant memory challenges, making low-rank adaptation methods an attractive solution. Previously, Low-Rank Adaptation (LoRA) addressed this by adding a trainable low-rank matrix to the frozen pre-trained weights in each layer, reducing the number of trainable parameters and optimizer states. GaLore, which compresses the gradient matrix instead of the weight matrix, has demonstrated superior performance to LoRA with faster convergence and reduced memory consumption. Despite their empirical success, the performance of these methods has not been fully understood or explained theoretically. In this paper, we analyze the optimization landscapes of LoRA, GaLore, and full-rank methods, revealing that GaLore benefits from fewer spurious local minima and a larger region that satisfies the \pl, a variant of Polyak-Łojasiewicz (PL) condition, leading to faster convergence. Our analysis leads to a novel method, GaRare, which further improves GaLore by using gradient random projection to reduce computational overhead. Practically, GaRare achieves strong performance in both pre-training and fine-tuning tasks, offering a more efficient approach to large-scale model adaptation.

Deep Learning · Graph Neural Networks

Qian Li, Minghui Ouyang, Tian Ding, Yuyi Wang, Qingjiang Shi, Ruoyu Sun

Graph neural networks (GNNs) have recently emerged as powerful tools for solving complex optimization problems, often being employed to approximate solution mappings. Empirical evidence shows that even shallow GNNs (with fewer than ten layers) can achieve strong performance in predicting optimal solutions to linear programming (LP) problems. This finding is somewhat counter-intuitive, as LPs are global optimization problems, while shallow GNNs predict based on local information. Although previous theoretical results suggest that GNNs have the expressive power to solve LPs, they require deep architectures whose depth grows at least polynomially with the problem size, and thus leave the underlying principle of this empirical phenomenon still unclear. In this paper, we examine this phenomenon through the lens of distributed computing and average-case analysis. We establish that the expressive power of GNNs for LPs is closely related to well-studied distributed algorithms for LPs. Specifically, we show that any $d$-round distributed LP algorithm can be simulated by a $d$-depth GNN, and vice versa. In particular, by designing a new distributed LP algorithm and then unrolling it, we prove that constant-depth, constant-width GNNs suffice to solve sparse binary LPs effectively. Here, in contrast with previous analyses focusing on worst-case scenarios, in which we show that GNN depth must increase with problem size by leveraging an impossibility result about distributed LP algorithms, our analysis shifts the focus to the average-case performance, and shows that constant GNN depth then becomes sufficient no matter how large the problem size is. Our theory is validated by numerical results.

Botao Ren, Xue Yang, Yi Yu, Junwei Luo, Zhidong Deng

Single point supervised oriented object detection has gained attention and made initial progress within the community. Diverse from those approaches relying on one-shot samples or powerful pretrained models (e.g. SAM), PointOBB has shown promise due to its prior-free feature. In this paper, we propose PointOBB-v2, a simpler, faster, and stronger method to generate pseudo rotated boxes from points without relying on any other prior. Specifically, we first generate a Class Probability Map (CPM) by training the network with non-uniform positive and negative sampling. We show that the CPM is able to learn the approximate object regions and their contours. Then, Principal Component Analysis (PCA) is applied to accurately estimate the orientation and the boundary of objects. By further incorporating a separation mechanism, we resolve the confusion caused by the overlapping on the CPM, enabling its operation in high-density scenarios. Extensive comparisons demonstrate that our method achieves a training speed 15.58$\times$ faster and an accuracy improvement of 11.60\%/25.15\%/21.19\% on the DOTA-v1.0/v1.5/v2.0 datasets compared to the previous state-of-the-art, PointOBB. This significantly advances the cutting edge of single point supervised oriented detection in the modular track. Code and models will be released.

Theory · Deep Learning

Binghao Liu, Han Yang, Fang Wan, Fei Gu

Deep learning has become essential in the biological species recognition task. However, a significant challenge is the ability to continuously learn new or mutated species with limited annotated samples. Since species within the same family typically share similar traits, distinguishing between new and existing (old) species during incremental learning often faces the issue of species confusion. This can result in "catastrophic forgetting" of old species and poor learning of new ones. To address this issue, we propose a Prototype Antithesis (PA) method, which leverages the hierarchical structures in biological taxa to reduce confusion between new and old species. PA operates in two steps: Residual Prototype Learning (RPL) and Residual Prototype Mixing (RPM). RPL enables the model to learn unique prototypes for each species alongside residual prototypes representing shared traits within families. RPM generates synthetic samples by blending features of new species with residual prototypes of old species, encouraging the model to focus on species-unique traits and minimize species confusion. By integrating RPL and RPM, the proposed PA method mitigates "catastrophic forgetting" while improving generalization to new species. Extensive experiments on CUB200, PlantVillage, and Tree-of-Life datasets demonstrate that PA significantly reduces inter-species confusion and achieves state-of-the-art performance, highlighting its potential for deep learning in biological data analysis.