CompCap: Improving Multimodal Large Language Models with Composite Captions

How well can Multimodal Large Language Models (MLLMs) understand composite images? Composite images (CIs) are synthetic visuals created by merging multiple visual elements, such as charts, posters, or screenshots, rather than being captured directly by a camera. While CIs are prevalent in real-world applications, recent MLLM developments have primarily focused on interpreting natural images (NIs). Our research reveals that current MLLMs face significant challenges in accurately understanding CIs, often struggling to extract information or perform complex reasoning based on these images. We find that existing training data for CIs are mostly formatted for question-answer tasks (e.g., in datasets like ChartQA and ScienceQA), while high-quality image-caption datasets, critical for robust vision-language alignment, are only available for NIs. To bridge this gap, we introduce Composite Captions (CompCap), a flexible framework that leverages Large Language Models (LLMs) and automation tools to synthesize CIs with accurate and detailed captions. Using CompCap, we curate CompCap-118K, a dataset containing 118K image-caption pairs across six CI types. We validate the effectiveness of CompCap-118K by supervised fine-tuning MLLMs of three sizes: xGen-MM-inst.-4B and LLaVA-NeXT-Vicuna-7B/13B. Empirical results show that CompCap-118K significantly enhances MLLMs' understanding of CIs, yielding average gains of 1.7%, 2.0%, and 2.9% across eleven benchmarks, respectively.

Adversarial-Robust Transfer Learning for Medical Imaging via Domain Assimilation

In the field of Medical Imaging, extensive research has been dedicated to leveraging its potential in uncovering critical diagnostic features in patients. Artificial Intelligence (AI)-driven medical diagnosis relies on sophisticated machine learning and deep learning models to analyze, detect, and identify diseases from medical images. Despite the remarkable performance of these models, characterized by high accuracy, they grapple with trustworthiness issues. The introduction of a subtle perturbation to the original image empowers adversaries to manipulate the prediction output, redirecting it to other targeted or untargeted classes. Furthermore, the scarcity of publicly available medical images, constituting a bottleneck for reliable training, has led contemporary algorithms to depend on pretrained models grounded on a large set of natural images -- a practice referred to as transfer learning. However, a significant {\em domain discrepancy} exists between natural and medical images, which causes AI models resulting from transfer learning to exhibit heightened {\em vulnerability} to adversarial attacks. This paper proposes a {\em domain assimilation} approach that introduces texture and color adaptation into transfer learning, followed by a texture preservation component to suppress undesired distortion. We systematically analyze the performance of transfer learning in the face of various adversarial attacks under different data modalities, with the overarching goal of fortifying the model's robustness and security in medical imaging tasks. The results demonstrate high effectiveness in reducing attack efficacy, contributing toward more trustworthy transfer learning in biomedical applications.

Does ChatGPT and Whisper Make Humanoid Robots More Relatable?

Humanoid robots are designed to be relatable to humans for applications such as customer support and helpdesk services. However, many such systems, including Softbank's Pepper, fall short because they fail to communicate effectively with humans. The advent of Large Language Models (LLMs) shows the potential to solve the communication barrier for humanoid robotics. This paper outlines the comparison of different Automatic Speech Recognition (ASR) APIs, the integration of Whisper ASR and ChatGPT with the Pepper robot and the evaluation of the system (Pepper-GPT) tested by 15 human users. The comparison result shows that, compared to the Google ASR and Google Cloud ASR, the Whisper ASR performed best as its average Word Error Rate (1.716%) and processing time (2.639 s) are both the lowest. The participants' usability investigations show that 60% of the participants thought the performance of the Pepper-GPT was "excellent", while the rest rated this system as "good" in the subsequent experiments. It is proved that while some problems still need to be overcome, such as the robot's multilingual ability and facial tracking capacity, users generally responded positively to the system, feeling like talking to an actual human.

Physics-informed Deep Learning to Solve Three-dimensional Terzaghi Consolidation Equation: Forward and Inverse Problems

The emergence of neural networks constrained by physical governing equations has sparked a new trend in deep learning research, which is known as Physics-Informed Neural Networks (PINNs). However, solving high-dimensional problems with PINNs is still a substantial challenge, the space complexity brings difficulty to solving large multidirectional problems. In this paper, a novel PINN framework to quickly predict several three-dimensional Terzaghi consolidation cases under different conditions is proposed. Meanwhile, the loss functions for different cases are introduced, and their differences in three-dimensional consolidation problems are highlighted. The tuning strategies for the PINNs framework for three-dimensional consolidation problems are introduced. Then, the performance of PINNs is tested and compared with traditional numerical methods adopted in forward problems, and the coefficients of consolidation and the impact of noisy data in inverse problems are identified. Finally, the results are summarized and presented from three-dimensional simulations of PINNs, which show an accuracy rate of over 99% compared with ground truth for both forward and inverse problems. These results are desirable with good accuracy and can be used for soil settlement prediction, which demonstrates that the proposed PINNs framework can learn the three-dimensional consolidation PDE well. Keywords: Three-dimensional Terzaghi consolidation; Physics-informed neural networks (PINNs); Forward problems; Inverse problems; soil settlement

A dynamic interactive learning framework for automated 3D medical image segmentation

Many deep learning based automated medical image segmentation systems, in reality, face difficulties in deployment due to the cost of massive data annotation and high latency in model iteration. We propose a dynamic interactive learning framework that addresses these challenges by integrating interactive segmentation into end-to-end weak supervised learning with streaming tasks. We develop novel replay and label smoothing schemes that overcome catastrophic forgetting and improve online learning robustness. For each image, our multi-round interactive segmentation module simultaneously optimizes both front-end predictions and deep learning segmenter. In each round, a 3D "proxy mask" is propagated from sparse user inputs based on image registration, serving as weak supervision that enable knowledge distillation from the unknown ground truth. In return, the trained segmenter explicitly guides next step's user interventions according to a spatial residual map from consecutive front or back-end predictions. Evaluation on 3D segmentation tasks (NCI-ISBI2013 and BraTS2015) shows that our framework generates online learning performances that match offline training benchmark. In addition, with a 62% reduction in total annotation efforts, our framework produces competitive dice scores comparing to online and offline learning which equipped with full ground truth. Furthermore, such a framework, with its flexibility and responsiveness, could be deployed behind hospital firewall that guarantees data security and easy maintenance.

Reason out Your Layout: Evoking the Layout Master from Large Language Models for Text-to-Image Synthesis

Recent advancements in text-to-image (T2I) generative models have shown remarkable capabilities in producing diverse and imaginative visuals based on text prompts. Despite the advancement, these diffusion models sometimes struggle to translate the semantic content from the text into images entirely. While conditioning on the layout has shown to be effective in improving the compositional ability of T2I diffusion models, they typically require manual layout input. In this work, we introduce a novel approach to improving T2I diffusion models using Large Language Models (LLMs) as layout generators. Our method leverages the Chain-of-Thought prompting of LLMs to interpret text and generate spatially reasonable object layouts. The generated layout is then used to enhance the generated images' composition and spatial accuracy. Moreover, we propose an efficient adapter based on a cross-attention mechanism, which explicitly integrates the layout information into the stable diffusion models. Our experiments demonstrate significant improvements in image quality and layout accuracy, showcasing the potential of LLMs in augmenting generative image models.

EDGE++: Improved Training and Sampling of EDGE

Recently developed deep neural models like NetGAN, CELL, and Variational Graph Autoencoders have made progress but face limitations in replicating key graph statistics on generating large graphs. Diffusion-based methods have emerged as promising alternatives, however, most of them present challenges in computational efficiency and generative performance. EDGE is effective at modeling large networks, but its current denoising approach can be inefficient, often leading to wasted computational resources and potential mismatches in its generation process. In this paper, we propose enhancements to the EDGE model to address these issues. Specifically, we introduce a degree-specific noise schedule that optimizes the number of active nodes at each timestep, significantly reducing memory consumption. Additionally, we present an improved sampling scheme that fine-tunes the generative process, allowing for better control over the similarity between the synthesized and the true network. Our experimental results demonstrate that the proposed modifications not only improve the efficiency but also enhance the accuracy of the generated graphs, offering a robust and scalable solution for graph generation tasks.

Generative quantum machine learning via denoising diffusion probabilistic models

Deep generative models are key-enabling technology to computer vision, text generation and large language models. Denoising diffusion probabilistic models (DDPMs) have recently gained much attention due to their ability to generate diverse and high-quality samples in many computer vision tasks, as well as to incorporate flexible model architectures and relatively simple training scheme. Quantum generative models, empowered by entanglement and superposition, have brought new insight to learning classical and quantum data. Inspired by the classical counterpart, we propose the quantum denoising diffusion probabilistic models (QuDDPM) to enable efficiently trainable generative learning of quantum data. QuDDPM adopts sufficient layers of circuits to guarantee expressivity, while introduces multiple intermediate training tasks as interpolation between the target distribution and noise to avoid barren plateau and guarantee efficient training. We demonstrate QuDDPM's capability in learning correlated quantum noise model and learning topological structure of nontrivial distribution of quantum data.

On Separate Normalization in Self-supervised Transformers

Self-supervised training methods for transformers have demonstrated remarkable performance across various domains. Previous transformer-based models, such as masked autoencoders (MAE), typically utilize a single normalization layer for both the [CLS] symbol and the tokens. We propose in this paper a simple modification that employs separate normalization layers for the tokens and the [CLS] symbol to better capture their distinct characteristics and enhance downstream task performance. Our method aims to alleviate the potential negative effects of using the same normalization statistics for both token types, which may not be optimally aligned with their individual roles. We empirically show that by utilizing a separate normalization layer, the [CLS] embeddings can better encode the global contextual information and are distributed more uniformly in its anisotropic space. When replacing the conventional normalization layer with the two separate layers, we observe an average 2.7% performance improvement over the image, natural language, and graph domains.

Catching Elusive Depression via Facial Micro-Expression Recognition

Depression is a common mental health disorder that can cause consequential symptoms with continuously depressed mood that leads to emotional distress. One category of depression is Concealed Depression, where patients intentionally or unintentionally hide their genuine emotions through exterior optimism, thereby complicating and delaying diagnosis and treatment and leading to unexpected suicides. In this paper, we propose to diagnose concealed depression by using facial micro-expressions (FMEs) to detect and recognize underlying true emotions. However, the extremely low intensity and subtle nature of FMEs make their recognition a tough task. We propose a facial landmark-based Region-of-Interest (ROI) approach to address the challenge, and describe a low-cost and privacy-preserving solution that enables self-diagnosis using portable mobile devices in a personal setting (e.g., at home). We present results and findings that validate our method, and discuss other technical challenges and future directions in applying such techniques to real clinical settings.