Multimodal Large Language Models (MLLMs) have showcased impressive skills in tasks related to visual understanding and reasoning. Yet, their widespread application faces obstacles due to the high computational demands during both the training and inference phases, restricting their use to a limited audience within the research and user communities. In this paper, we investigate the design aspects of Multimodal Small Language Models (MSLMs) and propose an efficient multimodal assistant named Mipha, which is designed to create synergy among various aspects: visual representation, language models, and optimization strategies. We show that without increasing the volume of training data, our Mipha-3B outperforms the state-of-the-art large MLLMs, especially LLaVA-1.5-13B, on multiple benchmarks. Through detailed discussion, we provide insights and guidelines for developing strong MSLMs that rival the capabilities of MLLMs. Our code is available at //github.com/zhuyiche/Mipha.
相關內容
Advanced Persistent Threats (APTs) represent the most threatening form of attack nowadays since they can stay undetected for a long time. Adversary emulation is a proactive approach for preparing against these attacks. However, adversary emulation tools lack the anti-detection abilities of APTs. We introduce Laccolith, a hypervisor-based solution for adversary emulation with anti-detection to fill this gap. We also present an experimental study to compare Laccolith with MITRE CALDERA, a state-of-the-art solution for adversary emulation, against five popular anti-virus products. We found that CALDERA cannot evade detection, limiting the realism of emulated attacks, even when combined with a state-of-the-art anti-detection framework. Our experiments show that Laccolith can hide its activities from all the tested anti-virus products, thus making it suitable for realistic emulations.
Robots can use Visual Imitation Learning (VIL) to learn everyday tasks from video demonstrations. However, translating visual observations into actionable robot policies is challenging due to the high-dimensional nature of video data. This challenge is further exacerbated by the morphological differences between humans and robots, especially when the video demonstrations feature humans performing tasks. To address these problems we introduce Visual Imitation lEarning with Waypoints (VIEW), an algorithm that significantly enhances the sample efficiency of human-to-robot VIL. VIEW achieves this efficiency using a multi-pronged approach: extracting a condensed prior trajectory that captures the demonstrator's intent, employing an agent-agnostic reward function for feedback on the robot's actions, and utilizing an exploration algorithm that efficiently samples around waypoints in the extracted trajectory. VIEW also segments the human trajectory into grasp and task phases to further accelerate learning efficiency. Through comprehensive simulations and real-world experiments, VIEW demonstrates improved performance compared to current state-of-the-art VIL methods. VIEW enables robots to learn a diverse range of manipulation tasks involving multiple objects from arbitrarily long video demonstrations. Additionally, it can learn standard manipulation tasks such as pushing or moving objects from a single video demonstration in under 30 minutes, with fewer than 20 real-world rollouts. Code and videos here: //collab.me.vt.edu/view/
The Effective Receptive field (ERF) plays an important role in transform coding, which determines how much redundancy can be removed at most during transform and how many spatial priors can be utilized to synthesize textures during inverse transform. Existing methods rely on stacks of small kernels, whose ERF remains not large enough instead, or heavy non-local attention mechanisms, which limit the potential of high-resolution image coding. To tackle this issue, we propose Large Receptive Field Transform Coding with Adaptive Weights for Learned Image Compression (LLIC). Specifically, for the first time in the learned image compression community, we introduce a few large kernel-based depth-wise convolutions to reduce more redundancy while maintaining modest complexity. Due to the wide range of image diversity, we further propose a mechanism to augment convolution adaptability through the self-conditioned generation of weights. The large kernels cooperate with non-linear embedding and gate mechanisms for better expressiveness and lighter point-wise interactions. Our investigation extends to refined training methods that unlock the full potential of these large kernels. Moreover, to promote more dynamic inter-channel interactions, we introduce an adaptive channel-wise bit allocation strategy that autonomously generates channel importance factors in a self-conditioned manner. To demonstrate the effectiveness of the proposed transform coding, we align the entropy model to compare with existing transform methods and obtain models LLIC-STF, LLIC-ELIC, LLIC-TCM. Extensive experiments demonstrate our proposed LLIC models have significant improvements over corresponding baselines and reduce BD-Rate by 9.49%, 9.47%, 10.94% on Kodak over VTM-17.0 Intra, respectively. Our LLIC models achieve state-of-the-art performances and better trade-offs between performance and complexity.
Despite the success of large language models (LLMs) in natural language generation, much evidence shows that LLMs may produce incorrect or nonsensical text. This limitation highlights the importance of discerning when to trust LLMs, especially in safety-critical domains. Existing methods, which rely on verbalizing confidence to tell the reliability by inducing top-k responses and sampling-aggregating multiple responses, often fail, due to the lack of objective guidance of confidence. To address this, we propose CONfidence-Quality-ORDerpreserving alignment approach (CONQORD), leveraging reinforcement learning with a tailored dual-component reward function. This function encompasses quality reward and orderpreserving alignment reward functions. Specifically, the order-preserving reward incentivizes the model to verbalize greater confidence for responses of higher quality to align the order of confidence and quality. Experiments demonstrate that our CONQORD significantly improves the alignment performance between confidence levels and response accuracy, without causing the model to become over-cautious. Furthermore, the aligned confidence provided by CONQORD informs when to trust LLMs, and acts as a determinant for initiating the retrieval process of external knowledge. Aligning confidence with response quality ensures more transparent and reliable responses, providing better trustworthiness.
Multi-modal foundation models such as CLIP have showcased impressive zero-shot capabilities. However, their applicability in resource-constrained environments is limited due to their large number of parameters and high inference time. While existing approaches have scaled down the entire CLIP architecture, we focus on training smaller variants of the image encoder, which suffices for efficient zero-shot classification. The use of synthetic data has shown promise in distilling representations from larger teachers, resulting in strong few-shot and linear probe performance. However, we find that this approach surprisingly fails in true zero-shot settings when using contrastive losses. We identify the exploitation of spurious features as being responsible for poor generalization between synthetic and real data. However, by using the image feature-based L2 distillation loss, we mitigate these problems and train students that achieve zero-shot performance which on four domain-specific datasets is on-par with a ViT-B/32 teacher model trained on DataCompXL, while featuring up to 92% fewer parameters.
Deepfake videos are becoming increasingly realistic, showing subtle tampering traces on facial areasthat vary between frames. Consequently, many existing Deepfake detection methods struggle to detect unknown domain Deepfake videos while accurately locating the tampered region. To address thislimitation, we propose Delocate, a novel Deepfake detection model that can both recognize andlocalize unknown domain Deepfake videos. Ourmethod consists of two stages named recoveringand localization. In the recovering stage, the modelrandomly masks regions of interest (ROIs) and reconstructs real faces without tampering traces, resulting in a relatively good recovery effect for realfaces and a poor recovery effect for fake faces. Inthe localization stage, the output of the recoveryphase and the forgery ground truth mask serve assupervision to guide the forgery localization process. This process strategically emphasizes the recovery phase of fake faces with poor recovery, facilitating the localization of tampered regions. Ourextensive experiments on four widely used benchmark datasets demonstrate that Delocate not onlyexcels in localizing tampered areas but also enhances cross-domain detection performance.
The rapidly evolving multimodal Large Language Models (LLMs) urgently require new benchmarks to uniformly evaluate their performance on understanding and textually describing music. However, due to semantic gaps between Music Information Retrieval (MIR) algorithms and human understanding, discrepancies between professionals and the public, and low precision of annotations, existing music description datasets cannot serve as benchmarks. To this end, we present MuChin, the first open-source music description benchmark in Chinese colloquial language, designed to evaluate the performance of multimodal LLMs in understanding and describing music. We established the Caichong Music Annotation Platform (CaiMAP) that employs an innovative multi-person, multi-stage assurance method, and recruited both amateurs and professionals to ensure the precision of annotations and alignment with popular semantics. Utilizing this method, we built a dataset with multi-dimensional, high-precision music annotations, the Caichong Music Dataset (CaiMD), and carefully selected 1,000 high-quality entries to serve as the test set for MuChin. Based on MuChin, we analyzed the discrepancies between professionals and amateurs in terms of music description, and empirically demonstrated the effectiveness of annotated data for fine-tuning LLMs. Ultimately, we employed MuChin to evaluate existing music understanding models on their ability to provide colloquial descriptions of music. All data related to the benchmark, along with the scoring code and detailed appendices, have been open-sourced (//github.com/CarlWangChina/MuChin/).
Recently, Segment Anything Model (SAM) shows exceptional performance in generating high-quality object masks and achieving zero-shot image segmentation. However, as a versatile vision model, SAM is primarily trained with large-scale natural light images. In underwater scenes, it exhibits substantial performance degradation due to the light scattering and absorption. Meanwhile, the simplicity of the SAM's decoder might lead to the loss of fine-grained object details. To address the above issues, we propose a novel feature learning framework named MAS-SAM for marine animal segmentation, which involves integrating effective adapters into the SAM's encoder and constructing a pyramidal decoder. More specifically, we first build a new SAM's encoder with effective adapters for underwater scenes. Then, we introduce a Hypermap Extraction Module (HEM) to generate multi-scale features for a comprehensive guidance. Finally, we propose a Progressive Prediction Decoder (PPD) to aggregate the multi-scale features and predict the final segmentation results. When grafting with the Fusion Attention Module (FAM), our method enables to extract richer marine information from global contextual cues to fine-grained local details. Extensive experiments on four public MAS datasets demonstrate that our MAS-SAM can obtain better results than other typical segmentation methods. The source code is available at //github.com/Drchip61/MAS-SAM.
Connecting text and visual modalities plays an essential role in generative intelligence. For this reason, inspired by the success of large language models, significant research efforts are being devoted to the development of Multimodal Large Language Models (MLLMs). These models can seamlessly integrate visual and textual modalities, both as input and output, while providing a dialogue-based interface and instruction-following capabilities. In this paper, we provide a comprehensive review of recent visual-based MLLMs, analyzing their architectural choices, multimodal alignment strategies, and training techniques. We also conduct a detailed analysis of these models across a wide range of tasks, including visual grounding, image generation and editing, visual understanding, and domain-specific applications. Additionally, we compile and describe training datasets and evaluation benchmarks, conducting comparisons among existing models in terms of performance and computational requirements. Overall, this survey offers a comprehensive overview of the current state of the art, laying the groundwork for future MLLMs.
Pre-trained Language Models (PLMs) have achieved great success in various Natural Language Processing (NLP) tasks under the pre-training and fine-tuning paradigm. With large quantities of parameters, PLMs are computation-intensive and resource-hungry. Hence, model pruning has been introduced to compress large-scale PLMs. However, most prior approaches only consider task-specific knowledge towards downstream tasks, but ignore the essential task-agnostic knowledge during pruning, which may cause catastrophic forgetting problem and lead to poor generalization ability. To maintain both task-agnostic and task-specific knowledge in our pruned model, we propose ContrAstive Pruning (CAP) under the paradigm of pre-training and fine-tuning. It is designed as a general framework, compatible with both structured and unstructured pruning. Unified in contrastive learning, CAP enables the pruned model to learn from the pre-trained model for task-agnostic knowledge, and fine-tuned model for task-specific knowledge. Besides, to better retain the performance of the pruned model, the snapshots (i.e., the intermediate models at each pruning iteration) also serve as effective supervisions for pruning. Our extensive experiments show that adopting CAP consistently yields significant improvements, especially in extremely high sparsity scenarios. With only 3% model parameters reserved (i.e., 97% sparsity), CAP successfully achieves 99.2% and 96.3% of the original BERT performance in QQP and MNLI tasks. In addition, our probing experiments demonstrate that the model pruned by CAP tends to achieve better generalization ability.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191