LLaVA-o1: Let Vision Language Models Reason Step-by-Step

Large language models have demonstrated substantial advancements in reasoning capabilities, particularly through inference-time scaling, as illustrated by models such as OpenAI's o1. However, current Vision-Language Models (VLMs) often struggle to perform systematic and structured reasoning, especially when handling complex visual question-answering tasks. In this work, we introduce LLaVA-o1, a novel VLM designed to conduct autonomous multistage reasoning. Unlike chain-of-thought prompting, LLaVA-o1 independently engages in sequential stages of summarization, visual interpretation, logical reasoning, and conclusion generation. This structured approach enables LLaVA-o1 to achieve marked improvements in precision on reasoning-intensive tasks. To accomplish this, we compile the LLaVA-o1-100k dataset, integrating samples from various visual question answering sources and providing structured reasoning annotations. Besides, we propose an inference-time stage-level beam search method, which enables effective inference-time scaling. Remarkably, with only 100k training samples and a simple yet effective inference time scaling method, LLaVA-o1 not only outperforms its base model by 8.9% on a wide range of multimodal reasoning benchmarks, but also surpasses the performance of larger and even closed-source models, such as Gemini-1.5-pro, GPT-4o-mini, and Llama-3.2-90B-Vision-Instruct.

Sparse Orthogonal Parameters Tuning for Continual Learning

Continual learning methods based on pre-trained models (PTM) have recently gained attention which adapt to successive downstream tasks without catastrophic forgetting. These methods typically refrain from updating the pre-trained parameters and instead employ additional adapters, prompts, and classifiers. In this paper, we from a novel perspective investigate the benefit of sparse orthogonal parameters for continual learning. We found that merging sparse orthogonality of models learned from multiple streaming tasks has great potential in addressing catastrophic forgetting. Leveraging this insight, we propose a novel yet effective method called SoTU (Sparse Orthogonal Parameters TUning). We hypothesize that the effectiveness of SoTU lies in the transformation of knowledge learned from multiple domains into the fusion of orthogonal delta parameters. Experimental evaluations on diverse CL benchmarks demonstrate the effectiveness of the proposed approach. Notably, SoTU achieves optimal feature representation for streaming data without necessitating complex classifier designs, making it a Plug-and-Play solution.

ETTFS: An Efficient Training Framework for Time-to-First-Spike Neuron

Spiking Neural Networks (SNNs) have attracted considerable attention due to their biologically inspired, event-driven nature, making them highly suitable for neuromorphic hardware. Time-to-First-Spike (TTFS) coding, where neurons fire only once during inference, offers the benefits of reduced spike counts, enhanced energy efficiency, and faster processing. However, SNNs employing TTFS coding often suffer from diminished classification accuracy. This paper presents an efficient training framework for TTFS that not only improves accuracy but also accelerates the training process. Unlike most previous approaches, we first identify two key issues limiting the performance of TTFS neurons: information disminishing and imbalanced membrane potential distribution. To address these challenges, we propose a novel initialization strategy. Additionally, we introduce a temporal weighting decoding method that aggregates temporal outputs through a weighted sum, supporting BPTT. Moreover, we re-evaluate the pooling layer in TTFS neurons and find that average pooling is better suited than max-pooling for this coding scheme. Our experimental results show that the proposed training framework leads to more stable training and significant performance improvements, achieving state-of-the-art (SOTA) results on both the MNIST and Fashion-MNIST datasets.

Spatial-Temporal Search for Spiking Neural Networks

Spiking Neural Networks (SNNs) are considered as a potential candidate for the next generation of artificial intelligence with appealing characteristics such as sparse computation and inherent temporal dynamics. By adopting architectures of Artificial Neural Networks (ANNs), SNNs achieve competitive performances on benchmark tasks like image classification. However, successful architectures of ANNs are not optimal for SNNs. In this work, we apply Neural Architecture Search (NAS) to find suitable architectures for SNNs. Previous NAS methods for SNNs focus primarily on the spatial dimension, with a notable lack of consideration for the temporal dynamics that are of critical importance for SNNs. Drawing inspiration from the heterogeneity of biological neural networks, we propose a differentiable approach to optimize SNN on both spatial and temporal dimensions. At spatial level, we have developed a spike-based differentiable hierarchical search (SpikeDHS) framework, where spike-based operation is optimized on both the cell and the layer level under computational constraints. We further propose a differentiable surrogate gradient search (DGS) method to evolve local SG functions independently during training. At temporal level, we explore an optimal configuration of diverse temporal dynamics on different types of spiking neurons by evolving their time constants, based on which we further develop hybrid networks combining SNN and ANN, balancing both accuracy and efficiency. Our methods achieve comparable classification performance of CIFAR10/100 and ImageNet with accuracies of 96.43%, 78.96%, and 70.21%, respectively. On event-based deep stereo, our methods find optimal layer variation and surpass the accuracy of specially designed ANNs with 26$\times$ lower computational cost ($6.7\mathrm{mJ}$), demonstrating the potential of SNN in processing highly sparse and dynamic signals.

Few-Shot Joint Multimodal Entity-Relation Extraction via Knowledge-Enhanced Cross-modal Prompt Model

Joint Multimodal Entity-Relation Extraction (JMERE) is a challenging task that aims to extract entities and their relations from text-image pairs in social media posts. Existing methods for JMERE require large amounts of labeled data. However, gathering and annotating fine-grained multimodal data for JMERE poses significant challenges. Initially, we construct diverse and comprehensive multimodal few-shot datasets fitted to the original data distribution. To address the insufficient information in the few-shot setting, we introduce the \textbf{K}nowledge-\textbf{E}nhanced \textbf{C}ross-modal \textbf{P}rompt \textbf{M}odel (KECPM) for JMERE. This method can effectively address the problem of insufficient information in the few-shot setting by guiding a large language model to generate supplementary background knowledge. Our proposed method comprises two stages: (1) a knowledge ingestion stage that dynamically formulates prompts based on semantic similarity guide ChatGPT generating relevant knowledge and employs self-reflection to refine the knowledge; (2) a knowledge-enhanced language model stage that merges the auxiliary knowledge with the original input and utilizes a transformer-based model to align with JMERE's required output format. We extensively evaluate our approach on a few-shot dataset derived from the JMERE dataset, demonstrating its superiority over strong baselines in terms of both micro and macro F$_1$ scores. Additionally, we present qualitative analyses and case studies to elucidate the effectiveness of our model.

MoH: Multi-Head Attention as Mixture-of-Head Attention

In this work, we upgrade the multi-head attention mechanism, the core of the Transformer model, to improve efficiency while maintaining or surpassing the previous accuracy level. We show that multi-head attention can be expressed in the summation form. Drawing on the insight that not all attention heads hold equal significance, we propose Mixture-of-Head attention (MoH), a new architecture that treats attention heads as experts in the Mixture-of-Experts (MoE) mechanism. MoH has two significant advantages: First, MoH enables each token to select the appropriate attention heads, enhancing inference efficiency without compromising accuracy or increasing the number of parameters. Second, MoH replaces the standard summation in multi-head attention with a weighted summation, introducing flexibility to the attention mechanism and unlocking extra performance potential. Extensive experiments on ViT, DiT, and LLMs demonstrate that MoH outperforms multi-head attention by using only 50%-90% of the attention heads. Moreover, we demonstrate that pre-trained multi-head attention models, such as LLaMA3-8B, can be further continue-tuned into our MoH models. Notably, MoH-LLaMA3-8B achieves an average accuracy of 64.0% across 14 benchmarks, outperforming LLaMA3-8B by 2.4% by utilizing only 75% of the attention heads. We believe the proposed MoH is a promising alternative to multi-head attention and provides a strong foundation for developing advanced and efficient attention-based models.

Is Parameter Collision Hindering Continual Learning in LLMs?

Large Language Models (LLMs) often suffer from catastrophic forgetting when learning multiple tasks sequentially, making continual learning (CL) essential for their dynamic deployment. Existing state-of-the-art (SOTA) methods, such as O-LoRA, typically focus on constructing orthogonality tasks to decouple parameter interdependence from various domains.In this paper, we reveal that building non-collision parameters is a more critical factor in addressing CL challenges. Our theoretical and experimental analyses demonstrate that non-collision parameters can provide better task orthogonality, which is a sufficient but unnecessary condition. Furthermore, knowledge from multiple domains will be preserved in non-collision parameter subspaces, making it more difficult to forget previously seen data. Leveraging this insight, we propose Non-collision Low-Rank Adaptation (N-LoRA), a simple yet effective approach leveraging low collision rates to enhance CL in LLMs. Experimental results on multiple CL benchmarks indicate that N-LoRA achieves superior performance (+2.9), higher task orthogonality (*4.1 times), and lower parameter collision (*58.1 times) than SOTA methods.

MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts

In this work, we aim to simultaneously enhance the effectiveness and efficiency of Mixture-of-Experts (MoE) methods. To achieve this, we propose MoE++, a general and heterogeneous MoE framework that integrates both Feed-Forward Network~(FFN) and zero-computation experts. Specifically, we introduce three types of zero-computation experts: the zero expert, copy expert, and constant expert, which correspond to discard, skip, and replace operations, respectively. This design offers three key advantages: (i) Low Computing Overhead: Unlike the uniform mixing mechanism for all tokens within vanilla MoE, MoE++ allows each token to engage with a dynamic number of FFNs, be adjusted by constant vectors, or even skip the MoE layer entirely. (ii) High Performance: By enabling simple tokens to utilize fewer FFN experts, MoE++ allows more experts to focus on challenging tokens, thereby unlocking greater performance potential than vanilla MoE. (iii) Deployment Friendly: Given that zero-computation experts have negligible parameters, we can deploy all zero-computation experts on each GPU, eliminating the significant communication overhead and expert load imbalance associated with FFN experts distributed across different GPUs. Moreover, we leverage gating residuals, enabling each token to consider the pathway taken in the previous layer when selecting the appropriate experts. Extensive experimental results demonstrate that MoE++ achieves better performance while delivering 1.1-2.1x expert forward throughput compared to a vanilla MoE model of the same size, which lays a solid foundation for developing advanced and efficient MoE-related models.

ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis

Despite recent advancements in neural 3D reconstruction, the dependence on dense multi-view captures restricts their broader applicability. In this work, we propose \textbf{ViewCrafter}, a novel method for synthesizing high-fidelity novel views of generic scenes from single or sparse images with the prior of video diffusion model. Our method takes advantage of the powerful generation capabilities of video diffusion model and the coarse 3D clues offered by point-based representation to generate high-quality video frames with precise camera pose control. To further enlarge the generation range of novel views, we tailored an iterative view synthesis strategy together with a camera trajectory planning algorithm to progressively extend the 3D clues and the areas covered by the novel views. With ViewCrafter, we can facilitate various applications, such as immersive experiences with real-time rendering by efficiently optimizing a 3D-GS representation using the reconstructed 3D points and the generated novel views, and scene-level text-to-3D generation for more imaginative content creation. Extensive experiments on diverse datasets demonstrate the strong generalization capability and superior performance of our method in synthesizing high-fidelity and consistent novel views.

OD-VAE: An Omni-dimensional Video Compressor for Improving Latent Video Diffusion Model

Variational Autoencoder (VAE), compressing videos into latent representations, is a crucial preceding component of Latent Video Diffusion Models (LVDMs). With the same reconstruction quality, the more sufficient the VAE's compression for videos is, the more efficient the LVDMs are. However, most LVDMs utilize 2D image VAE, whose compression for videos is only in the spatial dimension and often ignored in the temporal dimension. How to conduct temporal compression for videos in a VAE to obtain more concise latent representations while promising accurate reconstruction is seldom explored. To fill this gap, we propose an omni-dimension compression VAE, named OD-VAE, which can temporally and spatially compress videos. Although OD-VAE's more sufficient compression brings a great challenge to video reconstruction, it can still achieve high reconstructed accuracy by our fine design. To obtain a better trade-off between video reconstruction quality and compression speed, four variants of OD-VAE are introduced and analyzed. In addition, a novel tail initialization is designed to train OD-VAE more efficiently, and a novel inference strategy is proposed to enable OD-VAE to handle videos of arbitrary length with limited GPU memory. Comprehensive experiments on video reconstruction and LVDM-based video generation demonstrate the effectiveness and efficiency of our proposed methods.