EMMA: Empowering Multi-modal Mamba with Structural and Hierarchical Alignment

Mamba-based architectures have shown to be a promising new direction for deep learning models owing to their competitive performance and sub-quadratic deployment speed. However, current Mamba multi-modal large language models (MLLM) are insufficient in extracting visual features, leading to imbalanced cross-modal alignment between visual and textural latents, negatively impacting performance on multi-modal tasks. In this work, we propose Empowering Multi-modal Mamba with Structural and Hierarchical Alignment (EMMA), which enables the MLLM to extract fine-grained visual information. Specifically, we propose a pixel-wise alignment module to autoregressively optimize the learning and processing of spatial image-level features along with textual tokens, enabling structural alignment at the image level. In addition, to prevent the degradation of visual information during the cross-model alignment process, we propose a multi-scale feature fusion (MFF) module to combine multi-scale visual features from intermediate layers, enabling hierarchical alignment at the feature level. Extensive experiments are conducted across a variety of multi-modal benchmarks. Our model shows lower latency than other Mamba-based MLLMs and is nearly four times faster than transformer-based MLLMs of similar scale during inference. Due to better cross-modal alignment, our model exhibits lower degrees of hallucination and enhanced sensitivity to visual details, which manifests in superior performance across diverse multi-modal benchmarks. Code will be provided.

Blocks as Probes: Dissecting Categorization Ability of Large Multimodal Models

Categorization, a core cognitive ability in humans that organizes objects based on common features, is essential to cognitive science as well as computer vision. To evaluate the categorization ability of visual AI models, various proxy tasks on recognition from datasets to open world scenarios have been proposed. Recent development of Large Multimodal Models (LMMs) has demonstrated impressive results in high-level visual tasks, such as visual question answering, video temporal reasoning, etc., utilizing the advanced architectures and large-scale multimodal instruction tuning. Previous researchers have developed holistic benchmarks to measure the high-level visual capability of LMMs, but there is still a lack of pure and in-depth quantitative evaluation of the most fundamental categorization ability. According to the research on human cognitive process, categorization can be seen as including two parts: category learning and category use. Inspired by this, we propose a novel, challenging, and efficient benchmark based on composite blocks, called ComBo, which provides a disentangled evaluation framework and covers the entire categorization process from learning to use. By analyzing the results of multiple evaluation tasks, we find that although LMMs exhibit acceptable generalization ability in learning new categories, there are still gaps compared to humans in many ways, such as fine-grained perception of spatial relationship and abstract category understanding. Through the study of categorization, we can provide inspiration for the further development of LMMs in terms of interpretability and generalization.

GPSFormer: A Global Perception and Local Structure Fitting-based Transformer for Point Cloud Understanding

Despite the significant advancements in pre-training methods for point cloud understanding, directly capturing intricate shape information from irregular point clouds without reliance on external data remains a formidable challenge. To address this problem, we propose GPSFormer, an innovative Global Perception and Local Structure Fitting-based Transformer, which learns detailed shape information from point clouds with remarkable precision. The core of GPSFormer is the Global Perception Module (GPM) and the Local Structure Fitting Convolution (LSFConv). Specifically, GPM utilizes Adaptive Deformable Graph Convolution (ADGConv) to identify short-range dependencies among similar features in the feature space and employs Multi-Head Attention (MHA) to learn long-range dependencies across all positions within the feature space, ultimately enabling flexible learning of contextual representations. Inspired by Taylor series, we design LSFConv, which learns both low-order fundamental and high-order refinement information from explicitly encoded local geometric structures. Integrating the GPM and LSFConv as fundamental components, we construct GPSFormer, a cutting-edge Transformer that effectively captures global and local structures of point clouds. Extensive experiments validate GPSFormer's effectiveness in three point cloud tasks: shape classification, part segmentation, and few-shot learning. The code of GPSFormer is available at \url{https://github.com/changshuowang/GPSFormer}.

Q-Sparse: All Large Language Models can be Fully Sparsely-Activated

We introduce, Q-Sparse, a simple yet effective approach to training sparsely-activated large language models (LLMs). Q-Sparse enables full sparsity of activations in LLMs which can bring significant efficiency gains in inference. This is achieved by applying top-K sparsification to the activations and the straight-through-estimator to the training. The key results from this work are, (1) Q-Sparse can achieve results comparable to those of baseline LLMs while being much more efficient at inference time; (2) We present an inference-optimal scaling law for sparsely-activated LLMs; (3) Q-Sparse is effective in different settings, including training-from-scratch, continue-training of off-the-shelf LLMs, and finetuning; (4) Q-Sparse works for both full-precision and 1-bit LLMs (e.g., BitNet b1.58). Particularly, the synergy of BitNet b1.58 and Q-Sparse (can be equipped with MoE) provides the cornerstone and a clear path to revolutionize the efficiency, including cost and energy consumption, of future LLMs.

AIC MLLM: Autonomous Interactive Correction MLLM for Robust Robotic Manipulation

The ability to reflect on and correct failures is crucial for robotic systems to interact stably with real-life objects.Observing the generalization and reasoning capabilities of Multimodal Large Language Models (MLLMs), previous approaches have aimed to utilize these models to enhance robotic systems accordingly.However, these methods typically focus on high-level planning corrections using an additional MLLM, with limited utilization of failed samples to correct low-level contact poses. To address this gap, we propose an Autonomous Interactive Correction (AIC) MLLM, which makes use of previous low-level interaction experiences to correct SE(3) pose predictions. Specifically, AIC MLLM is initially fine-tuned to acquire both pose prediction and feedback prompt comprehension abilities.We carefully design two types of prompt instructions through interactions with objects: 1) visual masks to highlight unmovable parts for position correction, and 2)textual descriptions to indicate potential directions for rotation correction.During inference, a Feedback Information Extraction module is introduced to recognize the failure cause, allowing AIC MLLM to adaptively correct the pose prediction using the corresponding prompts.To further enhance manipulation stability, we devise a Test Time Adaptation strategy that enables AIC MLLM to better adapt to the current scene configuration.Finally, extensive experiments are conducted in both simulated and real-world environments to evaluate the proposed method. The results demonstrate that our AIC MLLM can efficiently correct failure samples by leveraging interaction experience prompts.Real-world demonstration can be found at https://sites.google.com/view/aic-mllm

M4U: Evaluating Multilingual Understanding and Reasoning for Large Multimodal Models

Multilingual multimodal reasoning is a core component in achieving human-level intelligence. However, most existing benchmarks for multilingual multimodal reasoning struggle to differentiate between models of varying performance; even language models without visual capabilities can easily achieve high scores. This leaves a comprehensive evaluation of leading multilingual multimodal models largely unexplored. In this work, we introduce M4U, a novel and challenging benchmark for assessing the capability of multi-discipline multilingual multimodal understanding and reasoning. M4U contains 8,931 samples covering 64 disciplines across 16 subfields in Science, Engineering, and Healthcare in Chinese, English, and German. Using M4U, we conduct extensive evaluations of 21 leading Large Multimodal Models (LMMs) and Large Language Models (LLMs) with external tools. The evaluation results show that the state-of-the-art model, GPT-4o, achieves only 47.6% average accuracy on M4U. Additionally, we observe that the leading LMMs exhibit significant language preferences. Our in-depth analysis indicates that leading LMMs, including GPT-4o, suffer performance degradation when prompted with cross-lingual multimodal questions, such as images with key textual information in Chinese while the question is in German. We believe that M4U can serve as a crucial tool for systematically evaluating LMMs based on their multilingual multimodal reasoning capabilities and monitoring their development. The homepage, codes and data are public available.

The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits

Recent research, such as BitNet, is paving the way for a new era of 1-bit Large Language Models (LLMs). In this work, we introduce a 1-bit LLM variant, namely BitNet b1.58, in which every single parameter (or weight) of the LLM is ternary {-1, 0, 1}. It matches the full-precision (i.e., FP16 or BF16) Transformer LLM with the same model size and training tokens in terms of both perplexity and end-task performance, while being significantly more cost-effective in terms of latency, memory, throughput, and energy consumption. More profoundly, the 1.58-bit LLM defines a new scaling law and recipe for training new generations of LLMs that are both high-performance and cost-effective. Furthermore, it enables a new computation paradigm and opens the door for designing specific hardware optimized for 1-bit LLMs.

Glance and Focus: Memory Prompting for Multi-Event Video Question Answering

Video Question Answering (VideoQA) has emerged as a vital tool to evaluate agents' ability to understand human daily behaviors. Despite the recent success of large vision language models in many multi-modal tasks, complex situation reasoning over videos involving multiple human-object interaction events still remains challenging. In contrast, humans can easily tackle it by using a series of episode memories as anchors to quickly locate question-related key moments for reasoning. To mimic this effective reasoning strategy, we propose the Glance-Focus model. One simple way is to apply an action detection model to predict a set of actions as key memories. However, these actions within a closed set vocabulary are hard to generalize to various video domains. Instead of that, we train an Encoder-Decoder to generate a set of dynamic event memories at the glancing stage. Apart from using supervised bipartite matching to obtain the event memories, we further design an unsupervised memory generation method to get rid of dependence on event annotations. Next, at the focusing stage, these event memories act as a bridge to establish the correlation between the questions with high-level event concepts and low-level lengthy video content. Given the question, the model first focuses on the generated key event memory, then focuses on the most relevant moment for reasoning through our designed multi-level cross-attention mechanism. We conduct extensive experiments on four Multi-Event VideoQA benchmarks including STAR, EgoTaskQA, AGQA, and NExT-QA. Our proposed model achieves state-of-the-art results, surpassing current large models in various challenging reasoning tasks. The code and models are available at https://github.com/ByZ0e/Glance-Focus.

BitNet: Scaling 1-bit Transformers for Large Language Models

The increasing size of large language models has posed challenges for deployment and raised concerns about environmental impact due to high energy consumption. In this work, we introduce BitNet, a scalable and stable 1-bit Transformer architecture designed for large language models. Specifically, we introduce BitLinear as a drop-in replacement of the nn.Linear layer in order to train 1-bit weights from scratch. Experimental results on language modeling show that BitNet achieves competitive performance while substantially reducing memory footprint and energy consumption, compared to state-of-the-art 8-bit quantization methods and FP16 Transformer baselines. Furthermore, BitNet exhibits a scaling law akin to full-precision Transformers, suggesting its potential for effective scaling to even larger language models while maintaining efficiency and performance benefits.

SEGA: Semantic Guided Attention on Visual Prototype for Few-Shot Learning

Teaching machines to recognize a new category based on few training samples especially only one remains challenging owing to the incomprehensive understanding of the novel category caused by the lack of data. However, human can learn new classes quickly even given few samples since human can tell what discriminative features should be focused on about each category based on both the visual and semantic prior knowledge. To better utilize those prior knowledge, we propose the SEmantic Guided Attention (SEGA) mechanism where the semantic knowledge is used to guide the visual perception in a top-down manner about what visual features should be paid attention to when distinguishing a category from the others. As a result, the embedding of the novel class even with few samples can be more discriminative. Concretely, a feature extractor is trained to embed few images of each novel class into a visual prototype with the help of transferring visual prior knowledge from base classes. Then we learn a network that maps semantic knowledge to category-specific attention vectors which will be used to perform feature selection to enhance the visual prototypes. Extensive experiments on miniImageNet, tieredImageNet, CIFAR-FS, and CUB indicate that our semantic guided attention realizes anticipated function and outperforms state-of-the-art results.