Kai Zhang

Zhiling Yan, Kai Zhang, Rong Zhou et al.

In this paper, we critically evaluate the capabilities of the state-of-the-art multimodal large language model, i.e., GPT-4 with Vision (GPT-4V), on Visual Question Answering (VQA) task. Our experiments thoroughly assess GPT-4V's proficiency in answering questions paired with images using both pathology and radiology datasets from 11 modalities (e.g. Microscopy, Dermoscopy, X-ray, CT, etc.) and fifteen objects of interests (brain, liver, lung, etc.). Our datasets encompass a comprehensive range of medical inquiries, including sixteen distinct question types. Throughout our evaluations, we devised textual prompts for GPT-4V, directing it to synergize visual and textual information. The experiments with accuracy score conclude that the current version of GPT-4V is not recommended for real-world diagnostics due to its unreliable and suboptimal accuracy in responding to diagnostic medical questions. In addition, we delineate seven unique facets of GPT-4V's behavior in medical VQA, highlighting its constraints within this complex arena. The complete details of our evaluation cases are accessible at https://github.com/ZhilingYan/GPT4V-Medical-Report.

Kai Zhang, John A. Lincoln, Xiaoqian Jiang et al.

Multiple Sclerosis (MS) is a chronic disease developed in human brain and spinal cord, which can cause permanent damage or deterioration of the nerves. The severity of MS disease is monitored by the Expanded Disability Status Scale (EDSS), composed of several functional sub-scores. Early and accurate classification of MS disease severity is critical for slowing down or preventing disease progression via applying early therapeutic intervention strategies. Recent advances in deep learning and the wide use of Electronic Health Records (EHR) creates opportunities to apply data-driven and predictive modeling tools for this goal. Previous studies focusing on using single-modal machine learning and deep learning algorithms were limited in terms of prediction accuracy due to the data insufficiency or model simplicity. In this paper, we proposed an idea of using patients' multimodal longitudinal and longitudinal EHR data to predict multiple sclerosis disease severity at the hospital visit. This work has two important contributions. First, we describe a pilot effort to leverage structured EHR data, neuroimaging data and clinical notes to build a multi-modal deep learning framework to predict patient's MS disease severity. The proposed pipeline demonstrates up to 25% increase in terms of the area under the Area Under the Receiver Operating Characteristic curve (AUROC) compared to models using single-modal data. Second, the study also provides insights regarding the amount useful signal embedded in each data modality with respect to MS disease prediction, which may improve data collection processes.

Zhenghao Pan, Haijin Zeng, Jiezhang Cao et al.

This paper endeavors to advance the precision of snapshot compressive imaging (SCI) reconstruction for multispectral image (MSI). To achieve this, we integrate the advantageous attributes of established SCI techniques and an image generative model, propose a novel structured zero-shot diffusion model, dubbed DiffSCI. DiffSCI leverages the structural insights from the deep prior and optimization-based methodologies, complemented by the generative capabilities offered by the contemporary denoising diffusion model. Specifically, firstly, we employ a pre-trained diffusion model, which has been trained on a substantial corpus of RGB images, as the generative denoiser within the Plug-and-Play framework for the first time. This integration allows for the successful completion of SCI reconstruction, especially in the case that current methods struggle to address effectively. Secondly, we systematically account for spectral band correlations and introduce a robust methodology to mitigate wavelength mismatch, thus enabling seamless adaptation of the RGB diffusion model to MSIs. Thirdly, an accelerated algorithm is implemented to expedite the resolution of the data subproblem. This augmentation not only accelerates the convergence rate but also elevates the quality of the reconstruction process. We present extensive testing to show that DiffSCI exhibits discernible performance enhancements over prevailing self-supervised and zero-shot approaches, surpassing even supervised transformer counterparts across both simulated and real datasets. Our code will be available.

Kai Zhang, Shan Liu, Momiao Xiong

A coordinate system is a foundation for every quantitative science, engineering, and medicine. Classical physics and statistics are based on the Cartesian coordinate system. The classical probability and hypothesis testing theory can only be applied to Euclidean data. However, modern data in the real world are from natural language processing, mathematical formulas, social networks, transportation and sensor networks, computer visions, automations, and biomedical measurements. The Euclidean assumption is not appropriate for non Euclidean data. This perspective addresses the urgent need to overcome those fundamental limitations and encourages extensions of classical probability theory and hypothesis testing , diffusion models and stochastic differential equations from Euclidean space to non Euclidean space. Artificial intelligence such as natural language processing, computer vision, graphical neural networks, manifold regression and inference theory, manifold learning, graph neural networks, compositional diffusion models for automatically compositional generations of concepts and demystifying machine learning systems, has been rapidly developed. Differential manifold theory is the mathematic foundations of deep learning and data science as well. We urgently need to shift the paradigm for data analysis from the classical Euclidean data analysis to both Euclidean and non Euclidean data analysis and develop more and more innovative methods for describing, estimating and inferring non Euclidean geometries of modern real datasets. A general framework for integrated analysis of both Euclidean and non Euclidean data, composite AI, decision intelligence and edge AI provide powerful innovative ideas and strategies for fundamentally advancing AI. We are expected to marry statistics with AI, develop a unified theory of modern statistics and drive next generation of AI and data science.

Bin Hong, Jinze Wu, Jiayu Liu et al.

In recent years, the breakthrough of Large Language Models (LLMs) offers new ideas for achieving universal methods on graph data. The common practice of converting graphs into natural language for LLMs, which refers to graph flattening, exhibits good generalizability and interpretability. However, the poor organization of the textual format results in poor performance in long-distance scenario understanding. Inspired by human cognitive reasoning habits, we propose a novel method for graph flattening to fit LLMs, termed as End-to-End DAG-Path prompting (EEDP). Experiments on real-world datasets show that EEDP enhances the reasoning performance of LLMs in long-distance scenarios while maintaining excellent performance in short-distance scenarios, demonstrating good robustness in the face of distance variations.

Ziqi Pang, Tianyuan Zhang, Fujun Luan et al.

We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences -- a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.

Tianyuan Zhang, Sai Bi, Yicong Hong et al.

Test-Time Training (TTT) models context dependencies by adapting part of the model's weights (referred to as fast weights) during inference. This fast weight, akin to recurrent states in RNNs, stores temporary memories of past tokens in the current sequence. Existing TTT methods struggled to show effectiveness in handling long-context data, due to their inefficiency on modern GPUs. The TTT layers in many of these approaches operate with extremely low FLOPs utilization (often <5%) because they deliberately apply small online minibatch sizes (e.g., updating fast weights every 16 or 64 tokens). Moreover, a small minibatch implies fine-grained block-wise causal dependencies in the data, unsuitable for data beyond 1D ordered sequences, like sets or N-dimensional grids such as images or videos. In contrast, we pursue the opposite direction by using an extremely large chunk update, ranging from 2K to 1M tokens across tasks of varying modalities, which we refer to as Large Chunk Test-Time Training (LaCT). It improves hardware utilization by orders of magnitude, and more importantly, facilitates scaling of nonlinear state size (up to 40% of model parameters), hence substantially improving state capacity, all without requiring cumbersome and error-prone kernel implementations. It also allows easy integration of sophisticated optimizers, e.g. Muon for online updates. We validate our approach across diverse modalities and tasks, including novel view synthesis with image set, language models, and auto-regressive video diffusion. Our approach can scale up to 14B-parameter AR video diffusion model on sequences up to 56K tokens. In our longest sequence experiment, we perform novel view synthesis with 1 million context length. We hope this work will inspire and accelerate new research in the field of long-context modeling and test-time training. Website: https://tianyuanzhang.com/projects/ttt-done-right

Yang Zheng, Hao Tan, Kai Zhang et al.

The rise of 3D Gaussian Splatting has revolutionized photorealistic 3D asset creation, yet a critical gap remains for their interactive refinement and editing. Existing approaches based on diffusion or optimization are ill-suited for this task, as they are often prohibitively slow, destructive to the original asset's identity, or lack the precision for fine-grained control. To address this, we introduce \ourmethod, a state-aware feedforward model that enables continuous editing of 3D Gaussian assets from user-provided 2D view(s). Our method directly predicts updates to the attributes of a compact, feature-rich Gaussian representation and leverages Test-Time Training to create a state-aware, iterative workflow. The versatility of our approach allows a single architecture to perform diverse tasks, including high-fidelity local detail refinement, local paint-over, and consistent global recoloring, all at interactive speeds, paving the way for fluid and intuitive 3D content authoring.

Kai Zhang, Siddharth Karanth, Bela Patel et al.

We propose a robust in-time predictor for in-hospital COVID-19 patient's probability of requiring mechanical ventilation. A challenge in the risk prediction for COVID-19 patients lies in the great variability and irregular sampling of patient's vitals and labs observed in the clinical setting. Existing methods have strong limitations in handling time-dependent features' complex dynamics, either oversimplifying temporal data with summary statistics that lose information or over-engineering features that lead to less robust outcomes. We propose a novel in-time risk trajectory predictive model to handle the irregular sampling rate in the data, which follows the dynamics of risk of performing mechanical ventilation for individual patients. The model incorporates the Multi-task Gaussian Process using observed values to learn the posterior joint multi-variant conditional probability and infer the missing values on a unified time grid. The temporal imputed data is fed into a multi-objective self-attention network for the prediction task. A novel positional encoding layer is proposed and added to the network for producing in-time predictions. The positional layer outputs a risk score at each user-defined time point during the entire hospital stay of an inpatient. We frame the prediction task into a multi-objective learning framework, and the risk scores at all time points are optimized altogether, which adds robustness and consistency to the risk score trajectory prediction. Our experimental evaluation on a large database with nationwide in-hospital patients with COVID-19 also demonstrates that it improved the state-of-the-art performance in terms of AUC (Area Under the receiver operating characteristic Curve) and AUPRC (Area Under the Precision-Recall Curve) performance metrics, especially at early times after hospital admission.