Visual language tasks such as Visual Question Answering (VQA) or Visual Entailment (VE) require AI models to comprehend and reason with both visual and textual content. Driven by the power of Large Language Models (LLMs), two prominent methods have emerged: (1) the hybrid integration between LLMs and Vision-Language Models (VLMs), where visual inputs are firstly converted into language descriptions by VLMs, serving as inputs for LLMs to generate final answer(s); (2) visual feature alignment in language space, where visual inputs are encoded as embeddings and projected to LLMs' language space via further supervised fine-tuning. The first approach provides light training costs and interpretability but is hard to be optimized in an end-to-end fashion. The second approach presents decent performance, but feature alignment usually requires large amounts of training data and lacks interpretability. To tackle this dilemma, we propose a novel approach, Inner Monologue Multi-Modal Optimization (IMMO), to solve complex vision language problems by simulating Inner Monologue, a cognitive process in which an individual engages in silent verbal communication with themselves. More specifically, we enable LLMs and VLMs to interact through natural language conversation (i.e., Inner Monologue) and propose to use a two-stage training process to learn how to do Inner Monologue (self-asking questions and answering questions). IMMO is evaluated on two popular tasks and achieves competitive performance with less training data when compared with state-of-the-art models while concurrently keeping the interpretability. The results suggest that by emulating the cognitive phenomenon of internal dialogue, our approach can enhance reasoning and explanation abilities, contributing to the more effective fusion of vision and language models. More importantly, instead of using predefined human-crafted monologues, IMMO learns this process within the deep learning models, broadening its potential applications across various AI challenges beyond vision and language tasks.

Although the Retrieval-Augmented Generation (RAG) paradigms can use external knowledge to enhance and ground the outputs of Large Language Models (LLMs) to mitigate generative hallucinations and static knowledge base problems, they still suffer from limited flexibility in adopting Information Retrieval (IR) systems with varying capabilities, constrained interpretability during the multi-round retrieval process, and a lack of end-to-end optimization. To address these challenges, we propose a novel LLM-centric approach, IM-RAG, that integrates IR systems with LLMs to support multi-round RAG through learning Inner Monologues (IM, i.e., the human inner voice that narrates one's thoughts). During the IM process, the LLM serves as the core reasoning model (i.e., Reasoner ) to either propose queries to collect more information via the Retriever or to provide a final answer based on the conversational context. We also introduce a Refiner that improves the outputs from the Retriever, effectively bridging the gap between the Reasoner and IR modules with varying capabilities and fostering multi-round communications. The entire IM process is optimized via Reinforcement Learning (RL) where a Progress Tracker is incorporated to provide mid-step rewards, and the answer prediction is further separately optimized via Supervised Fine-Tuning (SFT). We conduct extensive experiments with the HotPotQA dataset, a popular benchmark for retrieval-based, multi-step question-answering. The results show that our approach achieves state-of-the-art (SOTA) performance while providing high flexibility in integrating IR modules as well as strong interpretability exhibited in the learned inner monologue.

MOTIVATION: Morphological analyses with flatmount fluorescent images are essential to retinal pigment epithelial (RPE) aging studies and thus require accurate RPE cell segmentation. Although rapid technology advances in deep learning semantic segmentation have achieved great success in many biomedical research, the performance of these supervised learning methods for RPE cell segmentation is still limited by inadequate training data with high-quality annotations. RESULTS: To address this problem, we develop a Self-Supervised Semantic Segmentation (S4) method that utilizes a self-supervised learning strategy to train a semantic segmentation network with an encoder-decoder architecture. We employ a reconstruction and a pairwise representation loss to make the encoder extract structural information, while we create a morphology loss to produce the segmentation map. In addition, we develop a novel image augmentation algorithm (AugCut) to produce multiple views for self-supervised learning and enhance the network training performance. To validate the efficacy of our method, we applied our developed S4 method for RPE cell segmentation to a large set of flatmount fluorescent microscopy images, we compare our developed method for RPE cell segmentation with other state-of-the-art deep learning approaches. Compared with other state-of-the-art deep learning approaches, our method demonstrates better performance in both qualitative and quantitative evaluations, suggesting its promising potential to support large-scale cell morphological analyses in RPE aging investigations. AVAILABILITY AND IMPLEMENTATION: The codes and the documentation are available at: https://github.com/jkonglab/S4_RPE.

Automated curation of noisy external data in the medical domain has long been in high demand, as AI technologies need to be validated using various sources with clean, annotated data. Identifying the variance between internal and external sources is a fundamental step in curating a high-quality dataset, as the data distributions from different sources can vary significantly and subsequently affect the performance of AI models. The primary challenges for detecting data shifts are - (1) accessing private data across healthcare institutions for manual detection and (2) the lack of automated approaches to learn efficient shift-data representation without training samples. To overcome these problems, we propose an automated pipeline called MedShift to detect top-level shift samples and evaluate the significance of shift data without sharing data between internal and external organizations. MedShift employs unsupervised anomaly detectors to learn the internal distribution and identify samples showing significant shiftness for external datasets, and then compares their performance. To quantify the effects of detected shift data, we train a multi-class classifier that learns internal domain knowledge and evaluates the classification performance for each class in external domains after dropping the shift data. We also propose a data quality metric to quantify the dissimilarity between internal and external datasets. We verify the efficacy of MedShift using musculoskeletal radiographs (MURA) and chest X-ray datasets from multiple external sources. Our experiments show that our proposed shift data detection pipeline can be beneficial for medical centers to curate high-quality datasets more efficiently.

Language modality within the vision language pre-training framework is innately discretized, endowing each word in the language vocabulary a semantic meaning. In contrast, visual modality is inherently continuous and high-dimensional, which potentially prohibits the alignment as well as fusion between vision and language modalities. We therefore propose to "discretize" the visual representation by joint learning a codebook that imbues each visual token a semantic. We then utilize these discretized visual semantics as self-supervised ground-truths for building our Masked Image Modeling objective, a counterpart of Masked Language Modeling which proves successful for language models. To optimize the codebook, we extend the formulation of VQ-VAE which gives a theoretic guarantee. Experiments validate the effectiveness of our approach across common vision-language benchmarks.

Improving the retrieval relevance on noisy datasets is an emerging need for the curation of a large-scale clean dataset in the medical domain. While existing methods can be applied for class-wise retrieval (aka. inter-class), they cannot distinguish the granularity of likeness within the same class (aka. intra-class). The problem is exacerbated on medical external datasets, where noisy samples of the same class are treated equally during training. Our goal is to identify both intra/inter-class similarities for fine-grained retrieval. To achieve this, we propose an Outlier-Sensitive Content-based rAdiologhy Retrieval System (OSCARS), consisting of two steps. First, we train an outlier detector on a clean internal dataset in an unsupervised manner. Then we use the trained detector to generate the anomaly scores on the external dataset, whose distribution will be used to bin intra-class variations. Second, we propose a quadruplet (a, p, nintra, ninter) sampling strategy, where intra-class negatives nintra are sampled from bins of the same class other than the bin anchor a belongs to, while n_inter are randomly sampled from inter-classes. We suggest a weighted metric learning objective to balance the intra and inter-class feature learning. We experimented on two representative public radiography datasets. Experiments show the effectiveness of our approach. The training and evaluation code can be found in https://github.com/XiaoyuanGuo/oscars.

In this article, we focus on isolated gesture recognition and explore different modalities by involving RGB stream, depth stream, and saliency stream for inspection. Our goal is to push the boundary of this realm even further by proposing a unified framework that exploits the advantages of multi-modality fusion. Specifically, a spatial-temporal network architecture based on consensus-voting has been proposed to explicitly model the long-term structure of the video sequence and to reduce estimation variance when confronted with comprehensive inter-class variations. In addition, a three-dimensional depth-saliency convolutional network is aggregated in parallel to capture subtle motion characteristics. Extensive experiments are done to analyze the performance of each component and our proposed approach achieves the best results on two public benchmarks, ChaLearn IsoGD and RGBD-HuDaAct, outperforming the closest competitor by a margin of over 10% and 15%, respectively. Our project and codes will be released at https://davidsonic.github.io/index/acm_tomm_2017.html.

Path diversity exploits multiple routes simultaneously, achieving higher aggregated bandwidth and potentially decreasing delay and packet loss. Unfortunately, for TCP, naive load splitting often results in inaccurate estimation of round trip time (RTT) and packet reordering. As a result, it can suffer from significant instability or even throughput reduction. This is particular severe in Wireless Mesh Networks (WMNs), as validated by our analysis and simulation. To make multi-path TCP viable over WMNs, we propose a novel cross-layer design with a smart traffic split scheme, namely, Path Diversified Retransmission (PDR). PDR differentiates the original data packets and the retransmitted packets, and works with a novel QoS-aware multi-path routing protocol, QAOMDV, to distribute them separately. PDR does not suffer from the RTT underestimation and extra packet reordering, which ensures stable throughput improvement over single path routing. Through extensive simulations, we further demonstrate that, as compared to state-of-the-art multi-path protocols, our PDR with QAOMDV noticeably enhances the TCP throughput and reduces bandwidth fluctuation, with no obvious impact to fairness.