论文检索

Language

Silviu Pitis, Ziang Xiao, Nicolas Le Roux, Alessandro Sordoni

While finetuning language models from pairwise preferences has proven remarkably effective, the underspecified nature of natural language presents critical challenges. Direct preference feedback is uninterpretable, difficult to provide where multidimensional criteria may apply, and often inconsistent, either because it is based on incomplete instructions or provided by diverse principals. To address these challenges, we consider the two-step preference modeling procedure that first resolves the under-specification by selecting a context, and then evaluates preference with respect to the chosen context. We decompose reward modeling error according to these two steps, which suggests that supervising context in addition to context-specific preference may be a viable approach to aligning models with diverse human preferences. For this to work, the ability of models to evaluate context-specific preference is critical. To this end, we contribute context-conditioned preference datasets and accompanying experiments that investigate the ability of language models to evaluate context-specific preference. Unlike past datasets, where context-specific preference is highly correlated with general preference, our "preference reversal" datasets disentangle context-specific and general preferences to isolate context-specific capabilities. We use our datasets to (1) show that existing preference models benefit from, but fail to fully consider, added context, (2) finetune a context-aware reward model with context-specific performance exceeding that of GPT-4 and Llama 3 70B, and (3) investigate the potential value of context-aware preference modeling.

Zhihang Yuan, Hanling Zhang, Lu Pu, Xuefei Ning, Linfeng Zhang, Tianchen Zhao, Shengen Yan, Guohao Dai, Yu Wang

Diffusion Transformers (DiT) excel at image and video generation but face computational challenges due to the quadratic complexity of self-attention operators. We propose DiTFastAttn, a post-training compression method to alleviate the computational bottleneck of DiT.We identify three key redundancies in the attention computation during DiT inference: (1) spatial redundancy, where many attention heads focus on local information; (2) temporal redundancy, with high similarity between the attention outputs of neighboring steps; (3) conditional redundancy, where conditional and unconditional inferences exhibit significant similarity. We propose three techniques to reduce these redundancies: (1) $\textit{Window Attention with Residual Sharing}$ to reduce spatial redundancy; (2) $\textit{Attention Sharing across Timesteps}$ to exploit the similarity between steps; (3) $\textit{Attention Sharing across CFG}$ to skip redundant computations during conditional generation.

Reinforcement Learning · Function Approximation

Han-Dong Lim, Donghwan Lee

Q-learning is widely used algorithm in reinforcement learning (RL) community. Under the lookup table setting, its convergence is well established. However, its behavior is known to be unstable with the linear function approximation case. This paper develops a new Q-learning algorithm, called RegQ, that converges when linear function approximation is used. We prove that simply adding an appropriate regularization term ensures convergence of the algorithm. Its stability is established using a recent analysis tool based on switching system models. Moreover, we experimentally show that RegQ converges in environments where Q-learning with linear function approximation has known to diverge. An error bound on the solution where the algorithm converges is also given.

Computer Vision

Guozhen Zhang, Chuxnu Liu, Yutao Cui, Xiaotong Zhao, Kai Ma, Limin Wang

Inter-frame modeling is pivotal in generating intermediate frames for video frame interpolation (VFI). Current approaches predominantly rely on convolution or attention-based models, which often either lack sufficient receptive fields or entail significant computational overheads. Recently, Selective State Space Models (S6) have emerged, tailored specifically for long sequence modeling, offering both linear complexity and data-dependent modeling capabilities. In this paper, we propose VFIMamba, a novel frame interpolation method for efficient and dynamic inter-frame modeling by harnessing the S6 model. Our approach introduces the Mixed-SSM Block (MSB), which initially rearranges tokens from adjacent frames in an interleaved fashion and subsequently applies multi-directional S6 modeling. This design facilitates the efficient transmission of information across frames while upholding linear complexity. Furthermore, we introduce a novel curriculum learning strategy that progressively cultivates proficiency in modeling inter-frame dynamics across varying motion magnitudes, fully unleashing the potential of the S6 model. Experimental findings showcase that our method attains state-of-the-art performance across diverse benchmarks, particularly excelling in high-resolution scenarios. In particular, on the X-TEST dataset, VFIMamba demonstrates a noteworthy improvement of 0.80 dB for 4K frames and 0.96 dB for 2K frames.

Generative Models · Reasoning

Terufumi Morishita, Gaku Morio, Atsuki Yamaguchi, Yasuhiro Sogawa

Large language models (LLMs) are capable of solving a wide range of tasks, yet they have struggled with reasoning.To address this, we propose $\textbf{Additional Logic Training (ALT)}$, which aims to enhance LLMs' reasoning capabilities by program-generated logical reasoning samples.We first establish principles for designing high-quality samples by integrating symbolic logic theory and previous empirical insights.Then, based on these principles, we construct a synthetic corpus named $\textbf{Formal} \ \textbf{Logic} \ \textbf{\textit{D}eduction} \ \textbf{\textit{D}iverse}$ (FLD$ _{\times2}$), comprising numerous samples of multi-step deduction with unknown facts, diverse reasoning rules, diverse linguistic expressions, and challenging distractors.Finally, we empirically show that ALT on FLD$ _{\times2}$ substantially enhances the reasoning capabilities of state-of-the-art LLMs, including LLaMA-3.1-70B.Improvements include gains of up to 30 points on logical reasoning benchmarks, up to 10 points on math and coding benchmarks, and 5 points on the benchmark suite BBH.

Generative Models · Misc

Sahil Bhatia, Jie Qiu, Niranjan Hasabnis, Sanjit Seshia, Alvin Cheung

Domain-specific languages (DSLs) have become integral to various software workflows. Such languages offer domain-specific optimizations and abstractions that improve code readability and maintainability. However, leveraging these languages requires developers to rewrite existing code using the specific DSL's API. While large language models (LLMs) have shown some success in automatic code transpilation, none of them provide any functional correctness guarantees on the rewritten code. Another approach for automating this task is verified lifting, which relies on program synthesis to find programs in the target language that are functionally equivalent to the source language program. While several verified lifting tools have been developed for various application domains, they are specialized for specific source-target languages or require significant expertise in domain knowledge to make the search efficient. In this paper, leveraging recent advances in LLMs, we propose an LLM-based approach (LLMLift) to building verified lifting tools. We use the LLM's capabilities to reason about programs to translate a given program into its corresponding equivalent in the target language. Additionally, we use LLMs to generate proofs for functional equivalence. We develop lifting-based compilers for four DSLs targeting different application domains. Our approach not only outperforms previous symbolic-based tools in number of benchmarks transpiled and transpilation time, but also requires significantly less effort to build.

Reinforcement Learning · Everything Else

Yangru Huang, Peixi Peng, Yifan Zhao, Guangyao Chen, Yonghong Tian

Accurate environment dynamics modeling is crucial for obtaining effective state representations in visual reinforcement learning (RL) applications. However, when facing multiple input modalities, existing dynamics modeling methods (e.g., DeepMDP) usually stumble in addressing the complex and volatile relationship between different modalities. In this paper, we study the problem of efficient dynamics modeling for multi-modal visual RL. We find that under the existence of modality heterogeneity, modality-correlated and distinct features are equally important but play different roles in reflecting the evolution of environmental dynamics. Motivated by this fact, we propose Dissected Dynamics Modeling (DDM), a novel multi-modal dynamics modeling method for visual RL. Unlike existing methods, DDM explicitly distinguishes consistent and inconsistent information across modalities and treats them separately with a divide-and-conquer strategy. This is done by dispatching the features carrying different information into distinct dynamics modeling pathways, which naturally form a series of implicit regularizations along the learning trajectories. In addition, a reward predictive function is further introduced to filter task-irrelevant information in both modality-consistent and inconsistent features, ensuring information integrity while avoiding potential distractions. Extensive experiments show that DDM consistently achieves competitive performance in challenging multi-modal visual environments.

Deep Learning · Representation Learning

Beyazit Yalcinkaya, Niklas Lauffer, Marcell Vazquez-Chanlatte, Sanjit Seshia

Goal-conditioned reinforcement learning is a powerful way to control an AI agent's behavior at runtime. That said, popular goal representations, e.g., target states or natural language, are either limited to Markovian tasks or rely on ambiguous task semantics. We propose representing temporal goals using compositions of deterministic finite automata (cDFAs) and use cDFAs to guide RL agents. cDFAs balance the need for formal temporal semantics with ease of interpretation: if one can understand a flow chart, one can understand a cDFA. On the other hand, cDFAs form a countably infinite concept class with Boolean semantics, and subtle changes to the automaton can result in very different tasks, making them difficult to condition agent behavior on. To address this, we observe that all paths through a DFA correspond to a series of reach-avoid tasks and propose pre-training graph neural network embeddings on "reach-avoid derived" DFAs. Through empirical evaluation, we demonstrate that the proposed pre-training method enables zero-shot generalization to various cDFA task classes and accelerated policy specialization without the myopic suboptimality of hierarchical methods.

Computer Vision · Image Generation

Chenyi Zhuang, Ying Hu, Pan Gao

Text-to-image diffusion models particularly Stable Diffusion, have revolutionized the field of computer vision. However, the synthesis quality often deteriorates when asked to generate images that faithfully represent complex prompts involving multiple attributes and objects. While previous studies suggest that blended text embeddings lead to improper attribute binding, few have explored this in depth. In this work, we critically examine the limitations of the CLIP text encoder in understanding attributes and investigate how this affects diffusion models. We discern a phenomenon of attribute bias in the text space and highlight a contextual issue in padding embeddings that entangle different concepts. We propose Magnet, a novel training-free approach to tackle the attribute binding problem. We introduce positive and negative binding vectors to enhance disentanglement, further with a neighbor strategy to increase accuracy. Extensive experiments show that Magnet significantly improves synthesis quality and binding accuracy with negligible computational cost, enabling the generation of unconventional and unnatural concepts.

Robotics

Nan Song, Bozhou Zhang, Xiatian Zhu, Li Zhang

Motion forecasting for agents in autonomous driving is highly challenging due to the numerous possibilities for each agent's next action and their complex interactions in space and time. In real applications, motion forecasting takes place repeatedly and continuously as the self-driving car moves. However, existing forecasting methods typically process each driving scene within a certain range independently, totally ignoring the situational and contextual relationships between successive driving scenes. This significantly simplifies the forecasting task, making the solutions suboptimal and inefficient to use in practice. To address this fundamental limitation, we propose a novel motion forecasting framework for continuous driving, named RealMotion.It comprises two integral streams both at the scene level:(1) The scene context stream progressively accumulates historical scene information until the present moment, capturing temporal interactive relationships among scene elements.(2) The agent trajectory stream optimizes current forecasting by sequentially relaying past predictions.Besides, a data reorganization strategy is introduced to narrow the gap between existing benchmarks and real-world applications, consistent with our network. These approaches enable exploiting more broadly the situational and progressive insights of dynamic motion across space and time. Extensive experiments on Argoverse series with different settings demonstrate that our RealMotion achieves state-of-the-art performance, along with the advantage of efficient real-world inference.

Computer Vision · Segmentation

Bin-Bin Gao

Zero- and few-shot visual anomaly segmentation relies on powerful vision-language models that detect unseen anomalies using manually designed textual prompts. However, visual representations are inherently independent of language. In this paper, we explore the potential of a pure visual foundation model as an alternative to widely used vision-language models for universal visual anomaly segmentation.We present a novel paradigm that unifies anomaly segmentation into change segmentation. This paradigm enables us to leverage large-scale synthetic image pairs, featuring object-level and local region changes, derived from existing image datasets, which are independent of target anomaly datasets. We propose a one-prompt Meta-learning framework for Universal Anomaly Segmentation (MetaUAS) that is trained on this synthetic dataset and then generalizes well to segment any novel or unseen visual anomalies in the real world. To handle geometrical variations between prompt and query images, we propose a soft feature alignment module that bridges paired-image change perception and single-image semantic segmentation. This is the first work to achieve universal anomaly segmentation using a pure vision model without relying on special anomaly detection datasets and pre-trained visual-language models. Our method effectively and efficiently segments any anomalies with only one normal image prompt and enjoys training-free without guidance from language. Our MetaUAS significantly outperforms previous zero-shot, few-shot, and even full-shot anomaly segmentation methods. Code and Models: https://github.com/gaobb/MetaUAS.

Generative Models · In Context Learning

Zijian Zhou, Xiaoqiang Lin, Xinyi Xu, Alok Prakash, Daniela Rus, Bryan Kian Hsiang Low

In-context learning (ICL) allows transformer-based language models that are pre-trained on general text to quickly learn a specific task with a few "task demonstrations" without updating their parameters, significantly boosting their flexibility and generality. ICL possesses many distinct characteristics from conventional machine learning, thereby requiring new approaches to interpret this learning paradigm. Taking the viewpoint of recent works showing that transformers learn in context by formulating an internal optimizer, we propose an influence function-based attribution technique, DETAIL, that addresses the specific characteristics of ICL. We empirically verify the effectiveness of our approach for demonstration attribution while being computationally efficient. Leveraging the results, we then show how DETAIL can help improve model performance in real-world scenarios through demonstration reordering and curation. Finally, we experimentally prove the wide applicability of DETAIL by showing our attribution scores obtained on white-box models are transferable to black-box models in improving model performance.

Siyu Chen, Heejune Sheen, Tianhao Wang, Zhuoran Yang

In-context learning (ICL) is a cornerstone of large language model (LLM) functionality, yet its theoretical foundations remain elusive due to the complexity of transformer architectures. In particular, most existing work only theoretically explains how the attention mechanism facilitates ICL under certain data models. It remains unclear how the other building blocks of the transformer contribute to ICL. To address this question, we study how a two-attention-layer transformer is trained to perform ICL on $n$-gram Markov chain data, where each token in the Markov chain statistically depends on the previous n tokens. We analyze a sophisticated transformer model featuring relative positional embedding, multi-head softmax attention, and a feed-forward layer with normalization. We prove that the gradient flow with respect to a cross-entropy ICL loss converges to a limiting model that performs a generalized version of the "induction head" mechanism with a learned feature, resulting from the congruous contribution of all the building blocks. Specifically, the first attention layer acts as a copier, copying past tokens within a given window to each position, and the feed-forward network with normalization acts as a selector that generates a feature vector by only looking at informationally relevant parents from the window. Finally, the second attention layer is a classifier thatcompares these features with the feature at the output position, and uses the resulting similarity scores to generate the desired output. Our theory is further validated by simulation experiments.

Physical Models · Physics

Daniel Hothem, Ashe Miller, Timothy Proctor

Quantum computers have the potential to revolutionize diverse fields, including quantum chemistry, materials science, and machine learning. However, contemporary quantum computers experience errors that often cause quantum programs run on them to fail. Until quantum computers can reliably execute large quantum programs, stakeholders will need fast and reliable methods for assessing a quantum computer’s capability—i.e., the programs it can run and how well it can run them. Previously, off-the-shelf neural network architectures have been used to model quantum computers' capabilities, but with limited success, because these networks fail to learn the complex quantum physics that determines real quantum computers' errors. We address this shortcoming with a new quantum-physics-aware neural network architecture for learning capability models. Our scalable architecture combines aspects of graph neural networks with efficient approximations to the physics of errors in quantum programs. This approach achieves up to $\sim50\%$ reductions in mean absolute error on both experimental and simulated data, over state-of-the-art models based on convolutional neural networks, and scales to devices with 100+ qubits.

Computer Vision · Image Generation

Lianyu Pang, Jian Yin, Baoquan Zhao, Feize Wu, Fu Lee Wang, Qing Li, Xudong Mao

Recent advances in text-to-image models have enabled high-quality personalized image synthesis based on user-provided concepts with flexible textual control. In this work, we analyze the limitations of two primary techniques in text-to-image personalization: Textual Inversion and DreamBooth. When integrating the learned concept into new prompts, Textual Inversion tends to overfit the concept, while DreamBooth often overlooks it. We attribute these issues to the incorrect learning of the embedding alignment for the concept. To address this, we introduce AttnDreamBooth, a novel approach that separately learns the embedding alignment, the attention map, and the subject identity across different training stages. We also introduce a cross-attention map regularization term to enhance the learning of the attention map. Our method demonstrates significant improvements in identity preservation and text alignment compared to the baseline methods.

Ilias Diakonikolas, Nikos Zarifis

We study the problem of PAC learning $\gamma$-margin halfspaces in the presence of Massart noise. Without computational considerations, the sample complexity of this learning problem is known to be $\widetilde{\Theta}(1/(\gamma^2 \epsilon))$. Prior computationally efficient algorithms for the problem incur sample complexity $\tilde{O}(1/(\gamma^4 \epsilon^3))$ and achieve 0-1 error of $\eta+\epsilon$, where $\eta<1/2$ is the upper bound on the noise rate.Recent work gave evidence of an information-computation tradeoff, suggesting that a quadratic dependence on $1/\epsilon$ is required for computationally efficient algorithms. Our main result is a computationally efficient learner with sample complexity $\widetilde{\Theta}(1/(\gamma^2 \epsilon^2))$, nearly matching this lower bound. In addition, our algorithm is simple and practical, relying on online SGD on a carefully selected sequence of convex losses.

Generative Models · Reasoning

Xuezhi Wang, Denny Zhou

In enhancing the reasoning capabilities of large language models (LLMs), prior research primarily focuses on specific prompting techniques such as few-shot or zero-shot chain-of-thought (CoT) prompting. These methods, while effective, often involve manually intensive prompt engineering. Our study takes a novel approach by asking: Can LLMs reason effectively without any prompting? Our findings reveal that, intriguingly, CoT reasoning paths can be elicited from pre-trained LLMs by simply altering the \textit{decoding} process. Rather than conventional greedy decoding, we investigate the top-$k$ alternative tokens, uncovering that CoT paths are frequently inherent in these sequences. This approach not only bypasses the confounders of prompting but also allows us to assess the LLMs' \textit{intrinsic} reasoning abilities. Moreover, we observe that the presence of a CoT in the decoding path correlates with a higher confidence in the model's decoded answer. This confidence metric effectively differentiates between CoT and non-CoT paths. Extensive empirical studies on various reasoning benchmarks show that the proposed CoT-decoding effectively elicits reasoning capabilities from language models, which were previously obscured by standard greedy decoding.

Reinforcement Learning · Deep RL

Hao Bai, Yifei Zhou, Jiayi Pan, Mert Cemri, Alane Suhr, Sergey Levine, Aviral Kumar

Pre-trained vision language models (VLMs), though powerful, typically lack training on decision-centric data, rendering them sub-optimal for decision-making tasks such as in-the-wild device control through Graphical User Interfaces (GUIs) when used off-the-shelf. While training with static demonstrations has shown some promise, we show that such methods fall short when controlling real GUIs due to their failure to deal with real world stochasticity and dynamism not captured in static observational data. This paper introduces a novel autonomous RL approach, called DigiRL, for training in-the-wild device control agents through fine-tuning a pre-trained VLM in two stages: offline and offline-to-online RL. We first build a scalable and parallelizable Android learning environment equipped with a VLM-based general-purpose evaluator and then identify the key design choices for simple and effective RL in this domain. We demonstrate the effectiveness of DigiRL using the Android-in-the-Wild (AitW) dataset, where our 1.5B VLM trained with RL achieves a 49.5\% absolute improvement -- from 17.7 to 67.2\% success rate -- over supervised fine-tuning with static human demonstration data. It is worth noting that such improvement is achieved without any additional supervision or demonstration data. These results significantly surpass not only the prior best agents, including AppAgent with GPT-4V (8.3\% success rate) and the 17B CogAgent trained with AitW data (14.4\%), but also our implementation of prior best autonomous RL approach based on filtered behavior cloning (57.8\%), thereby establishing a new state-of-the-art for digital agents for in-the-wild device control.

Generative Models · Reasoning

Richard Yuanzhe Pang, Weizhe Yuan, He He, Kyunghyun Cho, Sainbayar Sukhbaatar, Jason Weston

Iterative preference optimization methods have recently been shown to perform well for general instruction tuning tasks, but typically make little improvement on reasoning tasks. In this work we develop an iterative approach that optimizes the preference between competing generated Chain-of-Thought (CoT) candidates by optimizing for winning vs. losing reasoning steps. We train using a modified DPO loss with an additional negative log-likelihood term, which we find to be crucial. We show reasoning improves across repeated iterations of this scheme. While only relying on examples in the training set, our approach results in increasing accuracy on GSM8K, MATH, and ARC-Challenge for Llama-2-70B-Chat, outperforming other Llama-2-based models not relying on additionally sourced datasets. For example, we see a large improvement from 55.6% to 81.6% on GSM8K and an accuracy of 88.7% with majority voting out of 32 samples.

Yanping Li, Jingshen Wang, Waverly Wei

Identifying subgroups with differential responses to treatment is pivotal in randomized clinical trials, as tailoring treatments to specific subgroups can advance personalized medicine. Upon trial completion, identifying best-performing subgroups–those with the most beneficial treatment effects–is crucial for optimizing resource allocation or mitigating adverse treatment effects. However, traditional clinical trials are not customized for the goal of identifying best-performing subgroups because they typically pre-define subgroups at the beginning of the trial and adhere to a fixed subgroup treatment allocation rule, leading to inefficient use of experimental efforts. While some adaptive experimental strategies exist for the identification of the single best subgroup, they commonly do not enable the identification of the best set of subgroups. To address these challenges, we propose a dynamic subgroup identification covariate-adjusted response-adaptive randomization (CARA) design strategy with the following key features: (i) Our approach is an adaptive experimental strategy that allows the dynamic identification of the best subgroups and the revision of treatment allocation towards the goal of correctly identifying the best subgroups based on collected experimental data. (ii) Our design handles ties between subgroups effectively, merging those with similar treatment effects to maximize experimental efficiency. In the theoretical investigations, we demonstrate that our design has a higher probability of correctly identifying the best set of subgroups compared to conventional designs. Additionally, we prove the statistical validity of our estimator for the best subgroup treatment effect, demonstrating its asymptotic normality and semiparametric efficiency. Finally, we validate our design using synthetic data from a clinical trial on cirrhosis.