Search

Large Language Models (LLMs) have achieved significant advancements, however, the common learning paradigm treats LLMs as passive information repositories, neglecting their potential for active learning and alignment. Some approaches train LLMs using their own generated synthetic data, exploring the possibility of active alignment. However, there is still a huge gap between these one-time alignment methods and the continuous automatic alignment of humans. In this paper, we introduce \textbf{I-SHEEP}, an \textbf{I}terative \textbf{S}elf-En\textbf{H}anc\textbf{E}m\textbf{E}nt \textbf{P}aradigm.This human-like paradigm enables LLMs to \textbf{continuously self-align from scratch with nothing}. Compared to the one-time alignment method Dromedary \cite{sun2023principledriven}, which refers to the first iteration in this paper, I-SHEEP can significantly enhance capacities on both Qwen and Llama models. I-SHEEP achieves a maximum relative improvement of 78.2\% in the Alpaca Eval, 24.0\% in the MT Bench, and an absolute increase of 8.88\% in the IFEval accuracy over subsequent iterations in Qwen-1.5 72B model. Additionally, I-SHEEP surpasses the base model in various standard benchmark generation tasks, achieving an average improvement of 24.77\% in code generation tasks, 12.04\% in TrivialQA, and 20.29\% in SQuAD. We also provide new insights based on the experiment results. Our codes, datasets, and models are available at \textbf{https://anonymous.4open.science/r/I-SHEEP}.

Diffusion Transformer, the backbone of Sora for video generation, successfully scales the capacity of diffusion models, pioneering new avenues for high-fidelity sequential data generation. Unlike static data such as images, sequential data consists of consecutive data frames indexed by time, exhibiting rich spatial and temporal dependencies. These dependencies represent the underlying dynamic model and are critical to validate the generated data. In this paper, we make the first theoretical step towards bridging diffusion transformers for capturing spatial-temporal dependencies. Specifically, we establish score approximation and distribution estimation guarantees of diffusion transformers for learning Gaussian process data with covariance functions of various decay patterns. We highlight how the spatial-temporal dependencies are captured and affect learning efficiency. Our study proposes a novel transformer approximation theory, where the transformer acts to unroll an algorithm. We support our theoretical results by numerical experiments, providing strong evidence that spatial-temporal dependencies are captured within attention layers, aligning with our approximation theory., Comment: 52 pages, 8 figures

Neural Radiance Fields (NeRF) are an advanced technology that creates highly realistic images by learning about scenes through a neural network model. However, NeRF often encounters issues when there are not enough images to work with, leading to problems in accurately rendering views. The main issue is that NeRF lacks sufficient structural details to guide the rendering process accurately. To address this, we proposed a Depth and Normal Dense Completion Priors for NeRF (CP\_NeRF) framework. This framework enhances view rendering by adding depth and normal dense completion priors to the NeRF optimization process. Before optimizing NeRF, we obtain sparse depth maps using the Structure from Motion (SfM) technique used to get camera poses. Based on the sparse depth maps and a normal estimator, we generate sparse normal maps for training a normal completion prior with precise standard deviations. During optimization, we apply depth and normal completion priors to transform sparse data into dense depth and normal maps with their standard deviations. We use these dense maps to guide ray sampling, assist distance sampling and construct a normal loss function for better training accuracy. To improve the rendering of NeRF's normal outputs, we incorporate an optical centre position embedder that helps synthesize more accurate normals through volume rendering. Additionally, we employ a normal patch matching technique to choose accurate rendered normal maps, ensuring more precise supervision for the model. Our method is superior to leading techniques in rendering detailed indoor scenes, even with limited input views.

Large Language Models (LLMs) for code are rapidly evolving, with code editing emerging as a critical capability. We introduce CodeEditorBench, an evaluation framework designed to rigorously assess the performance of LLMs in code editing tasks, including debugging, translating, polishing, and requirement switching. Unlike existing benchmarks focusing solely on code generation, CodeEditorBench emphasizes real-world scenarios and practical aspects of software development. We curate diverse coding challenges and scenarios from five sources, covering various programming languages, complexity levels, and editing tasks. Evaluation of 19 LLMs reveals that closed-source models (particularly Gemini-Ultra and GPT-4), outperform open-source models in CodeEditorBench, highlighting differences in model performance based on problem types and prompt sensitivities. CodeEditorBench aims to catalyze advancements in LLMs by providing a robust platform for assessing code editing capabilities. We will release all prompts and datasets to enable the community to expand the dataset and benchmark emerging LLMs. By introducing CodeEditorBench, we contribute to the advancement of LLMs in code editing and provide a valuable resource for researchers and practitioners.

The varying significance of distinct primitive behaviors during the policy learning process has been overlooked by prior model-free RL algorithms. Leveraging this insight, we explore the causal relationship between different action dimensions and rewards to evaluate the significance of various primitive behaviors during training. We introduce a causality-aware entropy term that effectively identifies and prioritizes actions with high potential impacts for efficient exploration. Furthermore, to prevent excessive focus on specific primitive behaviors, we analyze the gradient dormancy phenomenon and introduce a dormancy-guided reset mechanism to further enhance the efficacy of our method. Our proposed algorithm, ACE: Off-policy Actor-critic with Causality-aware Entropy regularization, demonstrates a substantial performance advantage across 29 diverse continuous control tasks spanning 7 domains compared to model-free RL baselines, which underscores the effectiveness, versatility, and efficient sample efficiency of our approach. Benchmark results and videos are available at https://ace-rl.github.io/., Comment: Accepted by ICML 2024

In this paper, we introduce Kun, a novel approach for creating high-quality instruction-tuning datasets for large language models (LLMs) without relying on manual annotations. Adapting a self-training algorithm based on instruction back-translation and answer polishment, Kun leverages unlabelled data from diverse sources such as Wudao, Wanjuan, and SkyPile to generate a substantial dataset of over a million Chinese instructional data points. This approach significantly deviates from traditional methods by using a self-curation process to refine and select the most effective instruction-output pairs. Our experiments with the 6B-parameter Yi model across various benchmarks demonstrate Kun's robustness and scalability. Our method's core contributions lie in its algorithmic advancement, which enhances data retention and clarity, and its innovative data generation approach that substantially reduces the reliance on costly and time-consuming manual annotations. This methodology presents a scalable and efficient solution for improving the instruction-following capabilities of LLMs, with significant implications for their application across diverse fields. The code and dataset can be found at https://github.com/Zheng0428/COIG-Kun, Comment: 12 pages, 12 figures

For multi-scale problems, the conventional physics-informed neural networks (PINNs) face some challenges in obtaining available predictions. In this paper, based on PINNs, we propose a practical deep learning framework for multi-scale problems by reconstructing the loss function and associating it with special neural network architectures. New PINN methods derived from the improved PINN framework differ from the conventional PINN method mainly in two aspects. First, the new methods use a novel loss function by modifying the standard loss function through a (grouping) regularization strategy. The regularization strategy implements a different power operation on each loss term so that all loss terms composing the loss function are of approximately the same order of magnitude, which makes all loss terms be optimized synchronously during the optimization process. Second, for the multi-frequency or high-frequency problems, in addition to using the modified loss function, new methods upgrade the neural network architecture from the common fully-connected neural network to special network architectures such as the Fourier feature architecture, and the integrated architecture developed by us. The combination of the above two techniques leads to a significant improvement in the computational accuracy of multi-scale problems. Several challenging numerical examples demonstrate the effectiveness of the proposed methods. The proposed methods not only significantly outperform the conventional PINN method in terms of computational efficiency and computational accuracy, but also compare favorably with the state-of-the-art methods in the recent literature. The improved PINN framework facilitates better application of PINNs to multi-scale problems.

Given the facts of the extensiveness of multi-material diffusion problems and the inability of the standard PINN(Physics-Informed Neural Networks) method for such problems, in this paper we present a novel PINN method that can accurately solve the multi-material diffusion equation. The new method applies continuity conditions at the material interface derived from the property of the diffusion equation, and combines the distinctive spatial separation strategy and the loss term normalization strategy to solve the problem that the residual points cannot be arranged at the material interface, the problem that it is difficult to express non-smooth functions with a single neural network, and the problem that the neural network is difficult to optimize the loss function with different magnitudes of loss terms, which finally provides the available prediction function for a class of multi-material diffusion problems. Numerical experiments verify the robustness and effectiveness of the new method.

The physics informed neural network (PINN) is a promising method for solving time-evolution partial differential equations (PDEs). However, the standard PINN method may fail to solve the PDEs with strongly nonlinear characteristics or those with high-frequency solutions. The physics informed neural network (PINN) is a promising method for solving time-evolution partial differential equations (PDEs). However, the standard PINN method may fail to solve the PDEs with strongly nonlinear characteristics or those with high-frequency solutions. The PT-PINN method transforms the difficult problem on the entire time domain to relatively simple problems defined on small subdomains. The neural network trained on small subdomains provides the neural network initialization and extra supervised learning data for the problems on larger subdomains or on the entire time-domain. By numerical experiments, we demonstrate that the PT-PINN succeeds in solving the evolution PDEs with strong non-linearity and/or high frequency solutions, including the strongly nonlinear heat equation, the Allen-Cahn equation, the convection equation with high-frequency solutions and so on, and that the convergence and accuracy of the PT-PINN is superior to the standard PINN method. The PT-PINN method is a competitive method for solving the time-evolution PDEs.