ICLR 2025 论文列表

Existing image super-resolution (SR) techniques often fail to generalize effectively in complex real-world settings due to the significant divergence between training data and practical scenarios. To address this challenge, previous efforts have either manually simulated intricate physical-based degradations or utilized learning-based techniques, yet these approaches remain inadequate for producing large-scale, realistic, and diverse data simultaneously. In this paper, we introduce a novel Realistic Decoupled Data Generator (RealDGen), an unsupervised learning data generation framework designed for real-world super-resolution. We meticulously develop content and degradation extraction strategies, which are integrated into a novel content-degradation decoupled diffusion model to create realistic low-resolution images from unpaired real LR and HR images. Extensive experiments demonstrate that RealDGen excels in generating large-scale, high-quality paired data that mirrors real-world degradations, significantly advancing the performance of popular SR models on various real-world benchmarks.

Augmenting LLMs with context leads to improved performance across many applications. Despite much research on Retrieval Augmented Generation (RAG) systems, an open question is whether errors arise because LLMs fail to utilize the context from retrieval or the context itself is insufficient to answer the query. To shed light on this, we develop a new notion of sufficient context, along with a method to classify instances that have enough information to answer the query. We then use sufficient context to analyze several models and datasets. By stratifying errors based on context sufficiency, we find that larger models with higher baseline performance (Gemini 1.5 Pro, GPT 4o, Claude 3.5) excel at answering queries when the context is sufficient, but often output incorrect answers instead of abstaining when the context is not. On the other hand, smaller models with lower baseline performance (Llama 3.1, Mistral 3, Gemma 2) hallucinate or abstain often, even with sufficient context. We further categorize cases when the context is useful, and improves accuracy, even though it does not fully answer the query and the model errs without the context. Building on our findings, we explore ways to reduce hallucinations in RAG systems, including a new selective generation method that leverages sufficient context information for guided abstention. Our method improves the fraction of correct answers among times where the model responds by 2--10% for Gemini, GPT, and Gemma. Code for our selective generation method and the prompts used in our autorater analysis are available on our [github](https://github.com/hljoren/sufficientcontext).

The predominant success of diffusion models in generative modeling has spurred significant interest in understanding their theoretical foundations. In this work, we propose a feature learning framework aimed at analyzing and comparing the training dynamics of diffusion models with those of traditional classification models. Our theoretical analysis demonstrates that diffusion models, due to the denoising objective, are encouraged to learn more balanced and comprehensive representations of the data. In contrast, neural networks with a similar architecture trained for classification tend to prioritize learning specific patterns in the data, often focusing on easy-to-learn components. To support these theoretical insights, we conduct several experiments on both synthetic and real-world datasets, which empirically validate our findings and highlight the distinct feature learning dynamics in diffusion models compared to classification.

Convolutional neural networks (CNNs) have been shown to be state-of-the-art models for visual cortical neurons. Cortical neurons in the primary visual cortex are sensitive to contextual information mediated by extensive horizontal and feedback connections. Standard CNNs integrate global contextual information to model contextual modulation via two mechanisms: successive convolutions and a fully connected readout layer. In this paper, we find that self-attention (SA), an implementation of non-local network mechanisms, can improve neural response predictions over parameter-matched CNNs in two key metrics: tuning curve correlation and peak tuning. We introduce peak tuning as a metric to evaluate a model's ability to capture a neuron's top feature preference. We factorize networks to assess each context mechanism, revealing that information in the local receptive field is most important for modeling overall tuning, but surround information is critically necessary for characterizing the tuning peak. We find that self-attention can replace posterior spatial-integration convolutions when learned incrementally, and is further enhanced in the presence of a fully connected readout layer, suggesting that the two context mechanisms are complementary. Finally, we find that decomposing receptive field learning and contextual modulation learning in an incremental manner may be an effective and robust mechanism for learning surround-center interactions.

We study the classical and parameterized complexity of computing the positive non-clashing teaching dimension of a set of concepts, that is, the smallest number of examples per concept required to successfully teach an intelligent learner under the considered, previously established model. For any class of concepts, it is known that this problem can be effortlessly transferred to the setting of balls in a graph $G$. We establish (1) the NP-hardness of the problem even when restricted to instances with positive non-clashing teaching dimension $k=2$ and where all balls in the graph are present, (2) near-tight running time upper and lower bounds for the problem on general graphs, (3) fixed-parameter tractability when parameterized by the vertex integrity of $G$, and (4) a lower bound excluding fixed-parameter tractability when parameterized by the feedback vertex number and pathwidth of $G$, even when combined with $k$. Our results provide a nearly complete understanding of the complexity landscape of computing the positive non-clashing teaching dimension and answer open questions from the literature.

3D Gaussian Splatting (3DGS) has witnessed its rapid development in novel view synthesis, which attains high quality reconstruction and real-time rendering. At the same time, there is still a gap before implicit neural representation (INR) can become a practical compressor due to the lack of stream decoding and real-time frame reconstruction on consumer-grade hardware. It remains a question whether the fast rendering and partial parameter decoding characteristics of 3DGS are applicable to video compression. To address these challenges, we propose a Toast-like Sliding Window (TSW) orthographic projection for converting any 3D Gaussian model into a video representation model. This method efficiently represents video by leveraging temporal redundancy through a sliding window approach. Additionally, the converted model is inherently stream-decodable and offers a higher rendering frame rate compared to INR methods. Building on TSW, we introduce an end-to-end trainable video compression method, GSVC, which employs deformable Gaussian representation and optical flow guidance to capture dynamic content in videos. Experimental results demonstrate that our method effectively transforms a 3D Gaussian model into a practical video compressor. GSVC further achieves better rate-distortion performance than NeRV on the UVG dataset, while achieving higher frame reconstruction speed (+30%~40% fps) and stream decoding. Code is available at [Github](https://github.com/actcwlf/GSVC)

Recent advances in Large Language Models (LLMs) have enabled the development of Video-LLMs, advancing multimodal learning by bridging video data with language tasks. However, current video understanding models struggle with processing long video sequences, supporting multi-turn dialogues, and adapting to real-world dynamic scenarios. To address these issues, we propose StreamChat, a training-free framework for streaming video reasoning and conversational interaction. StreamChat leverages a novel hierarchical memory system to efficiently process and compress video features over extended sequences, enabling real-time, multi-turn dialogue. Our framework incorporates a parallel system scheduling strategy that enhances processing speed and reduces latency, ensuring robust performance in real-world applications. Furthermore, we introduce StreamBench, a versatile benchmark that evaluates streaming video understanding across diverse media types and interactive scenarios, including multi-turn interactions and complex reasoning tasks. Extensive evaluations on StreamBench and other public benchmarks demonstrate that StreamChat significantly outperforms existing state-of-the-art models in terms of accuracy and response times, confirming its effectiveness for streaming video understanding. Code is available at StreamChat.

Consistent text-to-image generation depicting the *same* subjects across different images has gained significant recent attention due to its widespread applications in the fields of visual-storytelling and multiple-shot video generation. While remarkable, existing methods often require costly finetuning for each subject and struggle to maintain consistency across multiple characters. In this work, we first analyse the reason for these limitations. Our exploration reveals that the primary-issue stems from *self-attention leakage*, which is exacerbated when trying to ensure consistency across multiple-characters. Motivated by these findings, we next propose a simple yet effective *training and optimization-free approach* for improving multiple-character consistency. In particular, we first leverage multi-modal *chain-of-thought* reasoning in order to *apriori* localize the different subjects across the storyboard frames. The final storyboard images are then generated using a modified diffusion model which includes *1) a bounded cross-attention layer* for ensuring adherence to the initially predicted layout, and *2) a bounded cross-frame self-attention layer* for reducing inter-character attention leakage. Furthermore, we also propose a novel *cross-frame token-merging layer* which allows for improved fine-grain consistency for the storyboard characters. Experimental analysis reveals that proposed approach is not only $\times 30$ faster than prior training-based methods (*eg, textual inversion, dreambooth-lora*) but also surpasses the prior *state-of-the-art*, exhibiting improved multi-character consistency and text-to-image alignment performance.

Model-based methods in reinforcement learning offer a promising approach to enhance data efficiency by facilitating policy exploration within a dynamics model. However, accurately predicting sequential steps in the dynamics model remains a challenge due to the bootstrapping prediction, which attributes the next state to the prediction of the current state. This leads to accumulated errors during model roll-out. In this paper, we propose the Any-step Dynamics Model (ADM) to mitigate the compounding error by reducing bootstrapping prediction to direct prediction. ADM allows for the use of variable-length plans as inputs for predicting future states without frequent bootstrapping. We design two algorithms, ADMPO-ON and ADMPO-OFF, which apply ADM in online and offline model-based frameworks, respectively. In the online setting, ADMPO-ON demonstrates improved sample efficiency compared to previous state-of-the-art methods. In the offline setting, ADMPO-OFF not only demonstrates superior performance compared to recent state-of-the-art offline approaches but also offers better quantification of model uncertainty using only a single ADM.

Distributionally robust optimization (DRO) is a powerful technique to train robust machine learning models that perform well under distribution shifts. Compared with empirical risk minimization (ERM), DRO optimizes the expected loss under the worst-case distribution in an uncertainty set of distributions. This paper revisits the important problem of DRO with non-convex smooth loss functions. For this problem, Jin et al. (2021) showed that its dual problem is generalized $(L_0, L_1)$-smooth condition and gradient noise satisfies the affine variance condition, designed an algorithm of mini-batch normalized gradient descent with momentum, and proved its convergence and complexity. In this paper, we show that the dual problem and the gradient noise satisfy simpler yet more precise partially generalized smoothness condition and partially affine variance condition by studying the optimization variable and dual variable separately, which further yields much simpler algorithm design and convergence analysis. We develop a double stochastic gradient descent with clipping (D-SGD-C) algorithm that converges to an $\epsilon$-stationary point with $\mathcal O(\epsilon^{-4})$ gradient complexity, which matches with results in Jin et al. (2021). Our algorithm does not need to use momentum, and the proof is much simpler, thanks to the more precise characterization of partially generalized smoothness and partially affine variance noise. We further design a variance-reduced method that achieves a lower gradient complexity of $\mathcal O(\epsilon^{-3})$. Our theoretical results and insights are further verified numerically on a number of tasks, and our algorithms outperform the existing DRO method (Jin et al., 2021).

The performance and reasoning capabilities of Large Multi-modal Models (LMMs) is dependent on the size and quality of their training datasets. However, collecting datasets that support chain-of-thought instruction tuning is highly challenging. Existing video instruction tuning datasets are often derived by prompting large language models with video captions to generate question-answer pairs, which makes them predominantly descriptive rather than reasoning-focused. Meanwhile, many labeled video datasets with diverse labels and supervision exist -- however, we find that their integration into LMMs is non-trivial. Herein, we present $\underline{\text{Video}}$ $\underline{\text{S}}\text{elf}$-$\underline{\text{T}}\text{raining}$ $\text{with}$ $\underline{\text{a}}\text{ugmented}$ $\underline{\text{R}}\text{easoning}$ (Video-STaR), the first self-training approach for video instruction tuning. Video-STaR allows the utilization of *any* labeled video dataset for video instruction tuning. In Video-STaR, an LMM cycles between instruction generation and finetuning, which we show (I) improves general video understanding and (II) adapts LMMs to novel downstream tasks with existing supervision. During instruction generation, an LMM is prompted to propose an answer. The answers are then filtered only to those that contain the original video labels, and the LMM is then re-trained on the generated dataset. By training exclusively on generated answers containing the correct video labels, Video-STaR leverages these existing labels as weak supervision for video instruction tuning. Our results demonstrate that Video-STaR-augmented LMMs achieve notable improvements in (I) general Video QA, where TempCompass performance improved by 6.1%, *and* (II) downstream tasks, with a 9.9% increase in Kinetics700-QA accuracy and a 4.0% improvement in action quality assessment on FineDiving, while also exhibiting better interpretability.

Following its success for vision and text, the "foundation model" (FM) paradigm—pretraining large models on massive data, then fine-tuning on target tasks—has rapidly expanded to domains in the sciences, engineering, healthcare, and beyond. Has this achieved what the original FMs accomplished, i.e. the supplanting of traditional supervised learning in their domains? To answer we look at three modalities—genomics, satellite imaging, and time series—with multiple recent FMs and compare them to a standard supervised learning workflow: model development, hyperparameter tuning, and training, all using only data from the target task. Across these three specialized domains, we find that it is consistently possible to train simple supervised models—no more complicated than a lightly modified wide ResNet or UNet—that match or even outperform the latest foundation models. Our work demonstrates that the benefits of large-scale pretraining have yet to be realized in many specialized areas, reinforces the need to compare new FMs to strong, well-tuned baselines, and introduces two new, easy-to-use, open-source, and automated workflows for doing so.

The regularity or implicit bias in neural network optimization has been typically studied via the parameter norms or the landscape curvature, often overlooking the trajectory leading to these parameters. However, properties of the trajectory --- particularly its directionality --- capture critical aspects of how gradient descent navigates the landscape to converge to a solution. In this work, we introduce the notion of a Trajectory Map and derive natural complexity measures that highlight the directional characteristics of optimization trajectories. Our comprehensive analysis across vision and language modeling tasks reveals that (a) the trajectory's directionality at the macro-level saturates by the initial phase of training, wherein weight decay and momentum play a crucial but understated role; and (b) in subsequent training, trajectory directionality manifests in micro-level behaviors, such as oscillations, for which we also provide a theoretical analysis. This implies that neural optimization trajectories have, overall, a more linear form than zig-zaggy, as evident by high directional similarity, especially towards the end. To further hone this point, we show that when the trajectory direction gathers such an inertia, optimization proceeds largely unaltered even if the network is severely decapacitated (by freezing >99% of the parameters), --- thereby demonstrating the potential for significant computational and resource savings without compromising performance.

Multimodal visual language models are gaining prominence in open-world applications, driven by advancements in model architectures, training techniques, and high-quality data. However, their performance is often limited by insufficient task-specific data, leading to poor generalization and biased outputs. Existing efforts to increase task diversity in fine-tuning datasets are hindered by the labor-intensive process of manual task labeling, which typically produces only a few hundred task types. To address this, we propose TaskGalaxy, a large-scale multimodal instruction fine-tuning dataset comprising 19,227 hierarchical task types and 413,648 samples. TaskGalaxy utilizes GPT-4o to enrich task diversity by expanding from a small set of manually defined tasks, with CLIP and GPT-4o filtering those that best match open-source images, and generating relevant question-answer pairs. Multiple models are employed to ensure sample quality. This automated process enhances both task diversity and data quality, reducing manual intervention. Incorporating TaskGalaxy into LLaVA-v1.5 and InternVL-Chat-v1.0 models shows substantial performance improvements across 16 benchmarks, demonstrating the critical importance of task diversity. TaskGalaxy is publicly released at https://github.com/Kwai-YuanQi/TaskGalaxy.

Neural-network-based dynamics models learned from observational data have shown strong predictive capabilities for scene dynamics in robotic manipulation tasks. However, their inherent non-linearity presents significant challenges for effective planning. Current planning methods, often dependent on extensive sampling or local gradient descent, struggle with long-horizon motion planning tasks involving complex contact events. In this paper, we present a GPU-accelerated branch-and-bound (BaB) framework for motion planning in manipulation tasks that require trajectory optimization over neural dynamics models. Our approach employs a specialized branching heuristic to divide the search space into sub-domains and applies a modified bound propagation method, inspired by the state-of-the-art neural network verifier $\alpha,\beta$-CROWN, to efficiently estimate objective bounds within these sub-domains. The branching process guides planning effectively, while the bounding process strategically reduces the search space. Our framework achieves superior planning performance, generating high-quality state-action trajectories and surpassing existing methods in challenging, contact-rich manipulation tasks such as non-prehensile planar pushing with obstacles, object sorting, and rope routing in both simulated and real-world settings. Furthermore, our framework supports various neural network architectures, ranging from simple multilayer perceptrons to advanced graph neural dynamics models, and scales efficiently with different model sizes.

We propose practical deep Gaussian process models on Riemannian manifolds, similar in spirit to residual neural networks. With manifold-to-manifold hidden layers and an arbitrary last layer, they can model manifold- and scalar-valued functions, as well as vector fields. We target data inherently supported on manifolds, which is too complex for shallow Gaussian processes thereon. For example, while the latter perform well on high-altitude wind data, they struggle with the more intricate, nonstationary patterns at low altitudes. Our models significantly improve performance in these settings, enhancing prediction quality and uncertainty calibration, and remain robust to overfitting, reverting to shallow models when additional complexity is unneeded. We further showcase our models on Bayesian optimisation problems on manifolds, using stylised examples motivated by robotics, and obtain substantial improvements in later stages of the optimisation process. Finally, we show our models to have potential for speeding up inference for non-manifold data, when, and if, it can be mapped to a proxy manifold well enough.

Robotic manipulation in open-world settings requires not only task execution but also the ability to detect and learn from failures. While recent advances in vision-language models (VLMs) and large language models (LLMs) have improved robots' spatial reasoning and problem-solving abilities, they still struggle with failure recognition, limiting their real-world applicability. We introduce AHA, an open-source VLM designed to detect and reason about failures in robotic manipulation using natural language. By framing failure detection as a free-form reasoning task, AHA identifies failures and provides detailed, adaptable explanations across different robots, tasks, and environments. We fine-tuned AHA using FailGen, a scalable framework that generates the first large-scale dataset of robotic failure trajectories, the AHA dataset. FailGen achieves this by procedurally perturbing successful demonstrations from simulation. Despite being trained solely on the AHA dataset, AHA generalizes effectively to real-world failure datasets, robotic systems, and unseen tasks. It surpasses the second-best model (GPT-4o in-context learning) by 10.3% and exceeds the average performance of six compared models including five state-of-the-art VLMs by 35.3% across multiple metrics and datasets. We integrate AHA into three manipulation frameworks that utilize LLMs/VLMs for reinforcement learning, task and motion planning, and zero-shot trajectory generation. AHA’s failure feedback enhances these policies' performances by refining dense reward functions, optimizing task planning, and improving sub-task verification, boosting task success rates by an average of 21.4% across all three tasks compared to GPT-4 models. Project page: https://aha-vlm.github.io

Recent reasoning methods (e.g., chain-of-thought) help users understand how language models (LMs) answer a single question, but they do little to reveal the LM’s overall understanding, or “theory,” about the question’s topic, making it still hard to trust the model. Our goal is to materialize such theories - here called microtheories (a linguistic analog of logical microtheories) - as a set of sentences encapsulating an LM’s core knowledge about a topic. These statements systematically work together to entail answers to a set of questions to both engender trust and improve performance. Our approach is to first populate a knowledge store with (model-generated) sentences that entail answers to training questions, and then distill those down to a core microtheory which is concise, general, and non-redundant. We show that, when added to a general corpus (e.g., Wikipedia), microtheories can supply critical information not necessarily present in the corpus, improving both a model’s ability to ground its answers to verifiable knowledge (i.e., show how answers are systematically entailed by documents in the corpus, grounding up to +8% more answers), and the accuracy of those grounded answers (up to +8% absolute). We also show that, in a human evaluation in the medical domain, our distilled microtheories contain a significantly higher concentration of topically critical facts than the non-distilled knowledge store. Finally, we show we can quantify the coverage of a microtheory for a topic (characterized by a dataset) using a notion of p-relevance. Together, these suggest that microtheories are an efficient distillation of an LM’s topic-relevant knowledge, that they can usefully augment existing corpora, and can provide both performance gains and an interpretable, verifiable window into the model’s knowledge of a topic.

Large Vision-Language Models (VLMs) exhibit significant potential in multimodal tasks but often struggle with hallucinations—responses that are plausible yet visually ungrounded. In this work, we investigate the layer-wise prediction tendencies of VLMs and conduct an in-depth analysis of their decoding mechanism. We observe that VLMs tend to ``overthink'' during the final stages of decoding, making significant prediction shifts in the last few layers often favoring incorrect results, which leads to a surge in hallucinative outputs. Leveraging this localized pattern, we propose a novel decoding strategy inspired by the momentum analogy used in gradient descent-based optimizers. Our method enforces decoding consistency across layers in an adaptive manner during forward passes—an under-explored approach in existing works. This strategy significantly improves the reliability and performance of VLMs in various multimodal tasks, while introducing only negligible efficiency overhead.