The current state-of-the-art large language models (LLMs) are effective in generating high-quality text and encapsulating a broad spectrum of world knowledge. However, these models often hallucinate during generation and are not designed to utilize external information sources. To enable requests to the external knowledge bases, also called knowledge grounding, retrieval-augmented LLMs were introduced. For now, their applications have largely involved Open Domain Question Answering, Abstractive Question Answering, and such. In this paper, we broaden the scope of retrieval-augmented LLMs by venturing into a new task - code generation using external entities. For this task, we collect and publish a new dataset for project-level code generation, where the model should reuse functions defined in the project during generation. As we show, existing retrieval-augmented LLMs fail to assign relevance scores between similar entity names, and to mitigate it, they expand entity names with description context and append it to the input. In practice, due to the limited context size they can not accommodate the indefinitely large context of the whole project. To solve this issue, we propose a novel end-to-end trainable architecture with an scalable entity retriever injected directly into the LLM decoder. We demonstrate that our model can outperform common baselines in several scenarios, including project-level code generation, as well as Bash and SQL scripting.
We present GLEE in this work, an object-level foundation model for locating and identifying objects in images and videos. Through a unified framework, GLEE accomplishes detection, segmentation, tracking, grounding, and identification of arbitrary objects in the open world scenario for various object perception tasks. Adopting a cohesive learning strategy, GLEE acquires knowledge from diverse data sources with varying supervision levels to formulate general object representations, excelling in zero-shot transfer to new data and tasks. Specifically, we employ an image encoder, text encoder, and visual prompter to handle multi-modal inputs, enabling to simultaneously solve various object-centric downstream tasks while maintaining state-of-the-art performance. Demonstrated through extensive training on over five million images from diverse benchmarks, GLEE exhibits remarkable versatility and improved generalization performance, efficiently tackling downstream tasks without the need for task-specific adaptation. By integrating large volumes of automatically labeled data, we further enhance its zero-shot generalization capabilities. Additionally, GLEE is capable of being integrated into Large Language Models, serving as a foundational model to provide universal object-level information for multi-modal tasks. We hope that the versatility and universality of our method will mark a significant step in the development of efficient visual foundation models for AGI systems. The model and code will be released at https://glee-vision.github.io .
Diffusion models have exhibited impressive prowess in the text-to-image task. Recent methods add image-level controls, e.g., edge and depth maps, to manipulate the generation process together with text prompts to obtain desired images. This controlling process is globally operated on the entire image, which limits the flexibility of control regions. In this paper, we introduce a new simple yet practical task setting: local control. It focuses on controlling specific local areas according to user-defined image conditions, where the rest areas are only conditioned by the original text prompt. This manner allows the users to flexibly control the image generation in a fine-grained way. However, it is non-trivial to achieve this goal. The naive manner of directly adding local conditions may lead to the local control dominance problem. To mitigate this problem, we propose a training-free method that leverages the updates of noised latents and parameters in the cross-attention map during the denosing process to promote concept generation in non-control areas. Moreover, we use feature mask constraints to mitigate the degradation of synthesized image quality caused by information differences inside and outside the local control area. Extensive experiments demonstrate that our method can synthesize high-quality images to the prompt under local control conditions. Code is available at https://github.com/YibooZhao/Local-Control.
Transformers have astounding representational power but typically consume considerable computation which is quadratic with image resolution. The prevailing Swin transformer reduces computational costs through a local window strategy. However, this strategy inevitably causes two drawbacks: (1) the local window-based self-attention hinders global dependency modeling capability; (2) recent studies point out that local windows impair robustness. To overcome these challenges, we pursue a preferable trade-off between computational cost and performance. Accordingly, we propose a novel factorization self-attention mechanism (FaSA) that enjoys both the advantages of local window cost and long-range dependency modeling capability. By factorizing the conventional attention matrix into sparse sub-attention matrices, FaSA captures long-range dependencies while aggregating mixed-grained information at a computational cost equivalent to the local window-based self-attention. Leveraging FaSA, we present the factorization vision transformer (FaViT) with a hierarchical structure. FaViT achieves high performance and robustness, with linear computational complexity concerning input image spatial resolution. Extensive experiments have shown FaViT's advanced performance in classification and downstream tasks. Furthermore, it also exhibits strong model robustness to corrupted and biased data and hence demonstrates benefits in favor of practical applications. In comparison to the baseline model Swin-T, our FaViT-B2 significantly improves classification accuracy by 1% and robustness by 7%, while reducing model parameters by 14%. Our code will soon be publicly available at https://github.com/q2479036243/FaViT.
In the era of the Internet of Things (IoT), the retrieval of relevant medical information has become essential for efficient clinical decision-making. This paper introduces MedFusionRank, a novel approach to zero-shot medical information retrieval (MIR) that combines the strengths of pre-trained language models and statistical methods while addressing their limitations. The proposed approach leverages a pre-trained BERT-style model to extract compact yet informative keywords. These keywords are then enriched with domain knowledge by linking them to conceptual entities within a medical knowledge graph. Experimental evaluations on medical datasets demonstrate MedFusion Rank's superior performance over existing methods, with promising results with a variety of evaluation metrics. MedFusionRank demonstrates efficacy in retrieving relevant information, even from short or single-term queries.
The advent of increasingly powerful language models has raised expectations for language-based interactions. However, controlling these models is a challenge, emphasizing the need to be able to investigate the feasibility and value of their application. We present PROMISE, a framework that facilitates the development of complex language-based interactions with information systems. Its use of state machine modeling concepts enables model-driven, dynamic prompt orchestration across hierarchically nested states and transitions. This improves the control of the behavior of language models and thus enables their effective and efficient use. We show the benefits of PROMISE in the context of application scenarios within health information systems and demonstrate its ability to handle complex interactions.
In recent years, Artificial Intelligence Generated Content (AIGC) has gained widespread attention beyond the computer science community. Due to various issues arising from continuous creation of AI-generated images (AIGI), AIGC image quality assessment (AIGCIQA), which aims to evaluate the quality of AIGIs from human perception perspectives, has emerged as a novel topic in the field of computer vision. However, most existing AIGCIQA methods directly regress predicted scores from a single generated image, overlooking the inherent differences among AIGIs and scores. Additionally, operations like resizing and cropping may cause global geometric distortions and information loss, thus limiting the performance of models. To address these issues, we propose a patches sampling-based contrastive regression (PSCR) framework. We suggest introducing a contrastive regression framework to leverage differences among various generated images for learning a better representation space. In this space, differences and score rankings among images can be measured by their relative scores. By selecting exemplar AIGIs as references, we also overcome the limitations of previous models that could not utilize reference images on the no-reference image databases. To avoid geometric distortions and information loss in image inputs, we further propose a patches sampling strategy. To demonstrate the effectiveness of our proposed PSCR framework, we conduct extensive experiments on three mainstream AIGCIQA databases including AGIQA-1K, AGIQA-3K and AIGCIQA2023. The results show significant improvements in model performance with the introduction of our proposed PSCR framework. Code will be available at \url{https://github.com/jiquan123/PSCR}.
Despite the successful application of convolutional neural networks (CNNs) in object detection tasks, their efficiency in detecting faults from freight train images remains inadequate for implementation in real-world engineering scenarios. Existing modeling shortcomings of spatial invariance and pooling layers in conventional CNNs often ignore the neglect of crucial global information, resulting in error localization for fault objection tasks of freight trains. To solve these problems, we design a spatial-wise dynamic distillation framework based on multi-layer perceptron (MLP) for visual fault detection of freight trains. We initially present the axial shift strategy, which allows the MLP-like architecture to overcome the challenge of spatial invariance and effectively incorporate both local and global cues. We propose a dynamic distillation method without a pre-training teacher, including a dynamic teacher mechanism that can effectively eliminate the semantic discrepancy with the student model. Such an approach mines more abundant details from lower-level feature appearances and higher-level label semantics as the extra supervision signal, which utilizes efficient instance embedding to model the global spatial and semantic information. In addition, the proposed dynamic teacher can jointly train with students to further enhance the distillation efficiency. Extensive experiments executed on six typical fault datasets reveal that our approach outperforms the current state-of-the-art detectors and achieves the highest accuracy with real-time detection at a lower computational cost. The source code will be available at \url{https://github.com/MVME-HBUT/SDD-FTI-FDet}.
Due to the scarcity of publicly available diarization data, the model performance can be improved by training a single model with data from different domains. In this work, we propose to incorporate domain information to train a single end-to-end diarization model for multiple domains. First, we employ domain adaptive training with parameter-efficient adapters for on-the-fly model reconfiguration. Second, we introduce an auxiliary domain classification task to make the diarization model more domain-aware. For seen domains, the combination of our proposed methods reduces the absolute DER from 17.66% to 16.59% when compared with the baseline. During inference, adapters from ground-truth domains are not available for unseen domains. We demonstrate our model exhibits a stronger generalizability to unseen domains when adapters are removed. For two unseen domains, this improves the DER performance from 39.91% to 23.09% and 25.32% to 18.76% over the baseline, respectively.
The cross-domain capability of wireless sensing is currently one of the major challenges on human activity recognition (HAR) based on the channel state information (CSI) of wireless signals. The difficulty of labeling samples from new domains has encouraged the use of few and zero shot strategies. In this context, prototype networks have attracted attention due to their reasonable cross-domain transferability. This paper presents a novel zero-shot prototype recurrent convolutional network that implements a zero-shot learning strategy for HAR via CSI. This method extracts the prototypes from an available source domain to classify unseen and unlabeled data from the target domain for the same or similar classes. The experiments have been developed using three datasets with real measurements, and the results include an inter-datasets evaluation. Overall, the results improve the state of the art and make it a promising solution for cross-domain HAR.