Hao Zheng

Hao Zheng, Mirella Lapata

Compositional generalization is a basic mechanism in human language learning, which current neural networks struggle with. A recently proposed Disentangled sequence-to-sequence model (Dangle) shows promising generalization capability by learning specialized encodings for each decoding step. We introduce two key modifications to this model which encourage more disentangled representations and improve its compute and memory efficiency, allowing us to tackle compositional generalization in a more realistic setting. Specifically, instead of adaptively re-encoding source keys and values at each time step, we disentangle their representations and only re-encode keys periodically, at some interval. Our new architecture leads to better generalization performance across existing tasks and datasets, and a new machine translation benchmark which we create by detecting naturally occurring compositional patterns in relation to a training set. We show this methodology better emulates real-world requirements than artificial challenges.

Hao Zheng, Hui Lin, Rong Zhao et al.

The binding problem is one of the fundamental challenges that prevent the artificial neural network (ANNs) from a compositional understanding of the world like human perception, because disentangled and distributed representations of generative factors can interfere and lead to ambiguity when complex data with multiple objects are presented. In this paper, we propose a brain-inspired hybrid neural network (HNN) that introduces temporal binding theory originated from neuroscience into ANNs by integrating spike timing dynamics (via spiking neural networks, SNNs) with reconstructive attention (by ANNs). Spike timing provides an additional dimension for grouping, while reconstructive feedback coordinates the spikes into temporal coherent states. Through iterative interaction of ANN and SNN, the model continuously binds multiple objects at alternative synchronous firing times in the SNN coding space. The effectiveness of the model is evaluated on synthetic datasets of binary images. By visualization and analysis, we demonstrate that the binding is explainable, soft, flexible, and hierarchical. Notably, the model is trained on single object datasets without explicit supervision on grouping, but successfully binds multiple objects on test datasets, showing its compositional generalization capability. Further results show its binding ability in dynamic situations.

Bojian Hou, Hongming Li, Zhicheng Jiao et al.

Conventional survival analysis methods are typically ineffective to characterize heterogeneity in the population while such information can be used to assist predictive modeling. In this study, we propose a hybrid survival analysis method, referred to as deep clustering survival machines, that combines the discriminative and generative mechanisms. Similar to the mixture models, we assume that the timing information of survival data is generatively described by a mixture of certain numbers of parametric distributions, i.e., expert distributions. We learn weights of the expert distributions for individual instances according to their features discriminatively such that each instance's survival information can be characterized by a weighted combination of the learned constant expert distributions. This method also facilitates interpretable subgrouping/clustering of all instances according to their associated expert distributions. Extensive experiments on both real and synthetic datasets have demonstrated that the method is capable of obtaining promising clustering results and competitive time-to-event predicting performance.

Yejia Zhang, Pengfei Gu, Nishchal Sapkota et al.

High annotation costs and limited labels for dense 3D medical imaging tasks have recently motivated an assortment of 3D self-supervised pretraining methods that improve transfer learning performance. However, these methods commonly lack spatial awareness despite its centrality in enabling effective 3D image analysis. More specifically, position, scale, and orientation are not only informative but also automatically available when generating image crops for training. Yet, to date, no work has proposed a pretext task that distills all key spatial features. To fulfill this need, we develop a new self-supervised method, VectorPOSE, which promotes better spatial understanding with two novel pretext tasks: Vector Prediction (VP) and Boundary-Focused Reconstruction (BFR). VP focuses on global spatial concepts (i.e., properties of 3D patches) while BFR addresses weaknesses of recent reconstruction methods to learn more effective local representations. We evaluate VectorPOSE on three 3D medical image segmentation tasks, showing that it often outperforms state-of-the-art methods, especially in limited annotation settings.

Runqi Wang, Hao Zheng, Xiaoyue Duan et al.

Pre-trained vision-language models have inspired much research on few-shot learning. However, with only a few training images, there exist two crucial problems: (1) the visual feature distributions are easily distracted by class-irrelevant information in images, and (2) the alignment between the visual and language feature distributions is difficult. To deal with the distraction problem, we propose a Selective Attack module, which consists of trainable adapters that generate spatial attention maps of images to guide the attacks on class-irrelevant image areas. By messing up these areas, the critical features are captured and the visual distributions of image features are calibrated. To better align the visual and language feature distributions that describe the same object class, we propose a cross-modal distribution alignment module, in which we introduce a vision-language prototype for each class to align the distributions, and adopt the Earth Mover's Distance (EMD) to optimize the prototypes. For efficient computation, the upper bound of EMD is derived. In addition, we propose an augmentation strategy to increase the diversity of the images and the text prompts, which can reduce overfitting to the few-shot training images. Extensive experiments on 11 datasets demonstrate that our method consistently outperforms prior arts in few-shot learning. The implementation code will be available at https://github.com/bhrqw/SADA.

Teng Wang, Jinrui Zhang, Junjie Fei et al.

Controllable image captioning is an emerging multimodal topic that aims to describe the image with natural language following human purpose, $\textit{e.g.}$, looking at the specified regions or telling in a particular text style. State-of-the-art methods are trained on annotated pairs of input controls and output captions. However, the scarcity of such well-annotated multimodal data largely limits their usability and scalability for interactive AI systems. Leveraging unimodal instruction-following foundation models is a promising alternative that benefits from broader sources of data. In this paper, we present Caption AnyThing (CAT), a foundation model augmented image captioning framework supporting a wide range of multimodel controls: 1) visual controls, including points, boxes, and trajectories; 2) language controls, such as sentiment, length, language, and factuality. Powered by Segment Anything Model (SAM) and ChatGPT, we unify the visual and language prompts into a modularized framework, enabling the flexible combination between different controls. Extensive case studies demonstrate the user intention alignment capabilities of our framework, shedding light on effective user interaction modeling in vision-language applications. Our code is publicly available at https://github.com/ttengwang/Caption-Anything.

Bojun Zhang, Hangjian Ye, Hao Zheng et al.

Fine-grained 3D part segmentation is crucial for enabling embodied AI systems to perform complex manipulation tasks, such as interacting with specific functional components of an object. However, existing interactive segmentation methods are largely confined to coarse, instance-level targets, while non-interactive approaches struggle with sparse, real-world scans and suffer from a severe lack of annotated data. To address these limitations, we introduce PinPoint3D, a novel interactive framework for fine-grained, multi-granularity 3D segmentation, capable of generating precise part-level masks from only a few user point clicks. A key component of our work is a new 3D data synthesis pipeline that we developed to create a large-scale, scene-level dataset with dense part annotations, overcoming a critical bottleneck that has hindered progress in this field. Through comprehensive experiments and user studies, we demonstrate that our method significantly outperforms existing approaches, achieving an average IoU of around 55.8% on each object part under first-click settings and surpassing 71.3% IoU with only a few additional clicks. Compared to current state-of-the-art baselines, PinPoint3D yields up to a 16% improvement in IoU and precision, highlighting its effectiveness on challenging, sparse point clouds with high efficiency. Our work represents a significant step towards more nuanced and precise machine perception and interaction in complex 3D environments.

Fupei Guo, Hao Zheng, Xiang Zhang et al.

The rapid development of artificial intelligence has driven smart health with next-generation wireless communication technologies, stimulating exciting applications in remote diagnosis and intervention. To enable a timely and effective response for remote healthcare, efficient transmission of medical data through noisy channels with limited bandwidth emerges as a critical challenge. In this work, we propose a novel diffusion-based semantic communication framework, namely DiSC-Med, for the medical image transmission, where medical-enhanced compression and denoising blocks are developed for bandwidth efficiency and robustness, respectively. Unlike conventional pixel-wise communication framework, our proposed DiSC-Med is able to capture the key semantic information and achieve superior reconstruction performance with ultra-high bandwidth efficiency against noisy channels. Extensive experiments on real-world medical datasets validate the effectiveness of our framework, demonstrating its potential for robust and efficient telehealth applications.

Hao Zheng, Jinbao Wang, Xiantong Zhen et al.

Recently, Transformers have emerged as the go-to architecture for both vision and language modeling tasks, but their computational efficiency is limited by the length of the input sequence. To address this, several efficient variants of Transformers have been proposed to accelerate computation or reduce memory consumption while preserving performance. This paper presents an efficient vision Transformer, called CageViT, that is guided by convolutional activation to reduce computation. Our CageViT, unlike current Transformers, utilizes a new encoder to handle the rearranged tokens, bringing several technical contributions: 1) Convolutional activation is used to pre-process the token after patchifying the image to select and rearrange the major tokens and minor tokens, which substantially reduces the computation cost through an additional fusion layer. 2) Instead of using the class activation map of the convolutional model directly, we design a new weighted class activation to lower the model requirements. 3) To facilitate communication between major tokens and fusion tokens, Gated Linear SRA is proposed to further integrate fusion tokens into the attention mechanism. We perform a comprehensive validation of CageViT on the image classification challenge. Experimental results demonstrate that the proposed CageViT outperforms the most recent state-of-the-art backbones by a large margin in terms of efficiency, while maintaining a comparable level of accuracy (e.g. a moderate-sized 43.35M model trained solely on 224 x 224 ImageNet-1K can achieve Top-1 accuracy of 83.4% accuracy).

Hao Zheng, Mirella Lapata

There is mounting evidence that existing neural network models, in particular the very popular sequence-to-sequence architecture, struggle to systematically generalize to unseen compositions of seen components. We demonstrate that one of the reasons hindering compositional generalization relates to representations being entangled. We propose an extension to sequence-to-sequence models which encourages disentanglement by adaptively re-encoding (at each time step) the source input. Specifically, we condition the source representations on the newly decoded target context which makes it easier for the encoder to exploit specialized information for each prediction rather than capturing it all in a single forward pass. Experimental results on semantic parsing and machine translation empirically show that our proposal delivers more disentangled representations and better generalization.

Jie Liu, Hao Zheng

Due to the intractable partition function, the exact likelihood function for a Markov random field (MRF), in many situations, can only be approximated. Major approximation approaches include pseudolikelihood and Laplace approximation. In this paper, we propose a novel way of approximating the likelihood function through first approximating the marginal likelihood functions of individual parameters and then reconstructing the joint likelihood function from these marginal likelihood functions. For approximating the marginal likelihood functions, we derive a particular likelihood function from a modified scenario of coin tossing which is useful for capturing how one parameter interacts with the remaining parameters in the likelihood function. For reconstructing the joint likelihood function, we use an appropriate copula to link up these marginal likelihood functions. Numerical investigation suggests the superior performance of our approach. Especially as the size of the MRF increases, both the numerical performance and the computational cost of our approach remain consistently satisfactory, whereas Laplace approximation deteriorates and pseudolikelihood becomes computationally unbearable.