Kai Hu

Kai Hu, Andy Zou, Zifan Wang et al. · cmu

Despite the promise of Lipschitz-based methods for provably-robust deep learning with deterministic guarantees, current state-of-the-art results are limited to feed-forward Convolutional Networks (ConvNets) on low-dimensional data, such as CIFAR-10. This paper investigates strategies for expanding certifiably robust training to larger, deeper models. A key challenge in certifying deep networks is efficient calculation of the Lipschitz bound for residual blocks found in ResNet and ViT architectures. We show that fast ways of bounding the Lipschitz constant for conventional ResNets are loose, and show how to address this by designing a new residual block, leading to the \emph{Linear ResNet} (LiResNet) architecture. We then introduce \emph{Efficient Margin MAximization} (EMMA), a loss function that stabilizes robust training by simultaneously penalizing worst-case adversarial examples from \emph{all} classes. Together, these contributions yield new \emph{state-of-the-art} robust accuracy on CIFAR-10/100 and Tiny-ImageNet under $\ell_2$ perturbations. Moreover, for the first time, we are able to scale up fast deterministic robustness guarantees to ImageNet, demonstrating that this approach to robust learning can be applied to real-world applications. We release our code on Github: \url{https://github.com/klasleino/gloro}.

Zhihao Du, Shiliang Zhang, Kai Hu et al.

This paper presents FunCodec, a fundamental neural speech codec toolkit, which is an extension of the open-source speech processing toolkit FunASR. FunCodec provides reproducible training recipes and inference scripts for the latest neural speech codec models, such as SoundStream and Encodec. Thanks to the unified design with FunASR, FunCodec can be easily integrated into downstream tasks, such as speech recognition. Along with FunCodec, pre-trained models are also provided, which can be used for academic or generalized purposes. Based on the toolkit, we further propose the frequency-domain codec models, FreqCodec, which can achieve comparable speech quality with much lower computation and parameter complexity. Experimental results show that, under the same compression ratio, FunCodec can achieve better reconstruction quality compared with other toolkits and released models. We also demonstrate that the pre-trained models are suitable for downstream tasks, including automatic speech recognition and personalized text-to-speech synthesis. This toolkit is publicly available at https://github.com/alibaba-damo-academy/FunCodec.

Kai Hu, Klas Leino, Zifan Wang et al.

Recent studies have highlighted the potential of Lipschitz-based methods for training certifiably robust neural networks against adversarial attacks. A key challenge, supported both theoretically and empirically, is that robustness demands greater network capacity and more data than standard training. However, effectively adding capacity under stringent Lipschitz constraints has proven more difficult than it may seem, evident by the fact that state-of-the-art approach tend more towards \emph{underfitting} than overfitting. Moreover, we posit that a lack of careful exploration of the design space for Lipshitz-based approaches has left potential performance gains on the table. In this work, we provide a more comprehensive evaluation to better uncover the potential of Lipschitz-based certification methods. Using a combination of novel techniques, design optimizations, and synthesis of prior work, we are able to significantly improve the state-of-the-art VRA for deterministic certification on a variety of benchmark datasets, and over a range of perturbation sizes. Of particular note, we discover that the addition of large ``Cholesky-orthogonalized residual dense'' layers to the end of existing state-of-the-art Lipschitz-controlled ResNet architectures is especially effective for increasing network capacity and performance. Combined with filtered generative data augmentation, our final results further the state of the art deterministic VRA by up to 8.5 percentage points\footnote{Code is available at \url{https://github.com/hukkai/liresnet}}.

Xian Qian, Kai Hu, Jiaqiang Wang et al. · bytedance

This report describes our VolcTrans system for the WMT22 shared task on large-scale multilingual machine translation. We participated in the unconstrained track which allows the use of external resources. Our system is a transformerbased multilingual model trained on data from multiple sources including the public training set from the data track, NLLB data provided by Meta AI, self-collected parallel corpora, and pseudo bitext from back-translation. A series of heuristic rules clean both bilingual and monolingual texts. On the official test set, our system achieves 17.3 BLEU, 21.9 spBLEU, and 41.9 chrF2++ on average over all language pairs. The average inference speed is 11.5 sentences per second using a single Nvidia Tesla V100 GPU. Our code and trained models are available at https://github.com/xian8/wmt22

Zhihao Du, Qian Chen, Shiliang Zhang et al.

Recent years have witnessed a trend that large language model (LLM) based text-to-speech (TTS) emerges into the mainstream due to their high naturalness and zero-shot capacity. In this paradigm, speech signals are discretized into token sequences, which are modeled by an LLM with text as prompts and reconstructed by a token-based vocoder to waveforms. Obviously, speech tokens play a critical role in LLM-based TTS models. Current speech tokens are learned in an unsupervised manner, which lacks explicit semantic information and alignment to the text. In this paper, we propose to represent speech with supervised semantic tokens, which are derived from a multilingual speech recognition model by inserting vector quantization into the encoder. Based on the tokens, we further propose a scalable zero-shot TTS synthesizer, CosyVoice, which consists of an LLM for text-to-token generation and a conditional flow matching model for token-to-speech synthesis. Experimental results show that supervised semantic tokens significantly outperform existing unsupervised tokens in terms of content consistency and speaker similarity for zero-shot voice cloning. Moreover, we find that utilizing large-scale data further improves the synthesis performance, indicating the scalable capacity of CosyVoice. To the best of our knowledge, this is the first attempt to involve supervised speech tokens into TTS models.

Fangyi Chen, Han Zhang, Kai Hu et al.

This paper investigates a phenomenon where query-based object detectors mispredict at the last decoding stage while predicting correctly at an intermediate stage. We review the training process and attribute the overlooked phenomenon to two limitations: lack of training emphasis and cascading errors from decoding sequence. We design and present Selective Query Recollection (SQR), a simple and effective training strategy for query-based object detectors. It cumulatively collects intermediate queries as decoding stages go deeper and selectively forwards the queries to the downstream stages aside from the sequential structure. Such-wise, SQR places training emphasis on later stages and allows later stages to work with intermediate queries from earlier stages directly. SQR can be easily plugged into various query-based object detectors and significantly enhances their performance while leaving the inference pipeline unchanged. As a result, we apply SQR on Adamixer, DAB-DETR, and Deformable-DETR across various settings (backbone, number of queries, schedule) and consistently brings 1.4-2.8 AP improvement.

Ravi Mangal, Klas Leino, Zifan Wang et al. · cmu

Over the years, researchers have developed myriad attacks that exploit the ubiquity of adversarial examples, as well as defenses that aim to guard against the security vulnerabilities posed by such attacks. Of particular interest to this paper are defenses that provide provable guarantees against the class of $\ell_p$-bounded attacks. Certified defenses have made significant progress, taking robustness certification from toy models and datasets to large-scale problems like ImageNet classification. While this is undoubtedly an interesting academic problem, as the field has matured, its impact in practice remains unclear, thus we find it useful to revisit the motivation for continuing this line of research. There are three layers to this inquiry, which we address in this paper: (1) why do we care about robustness research? (2) why do we care about the $\ell_p$-bounded threat model? And (3) why do we care about certification as opposed to empirical defenses? In brief, we take the position that local robustness certification indeed confers practical value to the field of machine learning. We focus especially on the latter two questions from above. With respect to the first of the two, we argue that the $\ell_p$-bounded threat model acts as a minimal requirement for safe application of models in security-critical domains, while at the same time, evidence has mounted suggesting that local robustness may lead to downstream external benefits not immediately related to robustness. As for the second, we argue that (i) certification provides a resolution to the cat-and-mouse game of adversarial attacks; and furthermore, that (ii) perhaps contrary to popular belief, there may not exist a fundamental trade-off between accuracy, robustness, and certifiability, while moreover, certified training techniques constitute a particularly promising way for learning robust models.

Zhihao Du, Jiaming Wang, Qian Chen et al.

Generative Pre-trained Transformer (GPT) models have achieved remarkable performance on various natural language processing tasks, and have shown great potential as backbones for audio-and-text large language models (LLMs). Previous mainstream audio-and-text LLMs use discrete audio tokens to represent both input and output audio; however, they suffer from performance degradation on tasks such as automatic speech recognition, speech-to-text translation, and speech enhancement over models using continuous speech features. In this paper, we propose LauraGPT, a novel unified audio-and-text GPT-based LLM for audio recognition, understanding, and generation. LauraGPT is a versatile LLM that can process both audio and text inputs and generate outputs in either modalities. We propose a novel data representation that combines continuous and discrete features for audio: LauraGPT encodes input audio into continuous representations using an audio encoder and generates output audio from discrete codec codes. We propose a one-step codec vocoder to overcome the prediction challenge caused by the multimodal distribution of codec tokens. We fine-tune LauraGPT using supervised multi-task learning. Extensive experiments show that LauraGPT consistently achieves comparable to superior performance compared to strong baselines on a wide range of audio tasks related to content, semantics, paralinguistics, and audio-signal analysis, such as automatic speech recognition, speech-to-text translation, text-to-speech synthesis, speech enhancement, automated audio captioning, speech emotion recognition, and spoken language understanding.

Yansong Li, Kai Hu, Kohei Nakajima et al.

Echo state network (ESN), a kind of recurrent neural networks, consists of a fixed reservoir in which neurons are connected randomly and recursively and obtains the desired output only by training output connection weights. First-order reduced and controlled error (FORCE) learning is an online supervised training approach that can change the chaotic activity of ESNs into specified activity patterns. This paper proposes a composite FORCE learning method based on recursive least squares to train ESNs whose initial activity is spontaneously chaotic, where a composite learning technique featured by dynamic regressor extension and memory data exploitation is applied to enhance parameter convergence. The proposed method is applied to a benchmark problem about predicting chaotic time series generated by the Mackey-Glass system, and numerical results have shown that it significantly improves learning and prediction performances compared with existing methods.

Kai Hu, Zhuoyuan Wu, Zhuoyao Zhong et al.

In this paper, we present a new question-answering (QA) based key-value pair extraction approach, called KVPFormer, to robustly extracting key-value relationships between entities from form-like document images. Specifically, KVPFormer first identifies key entities from all entities in an image with a Transformer encoder, then takes these key entities as \textbf{questions} and feeds them into a Transformer decoder to predict their corresponding \textbf{answers} (i.e., value entities) in parallel. To achieve higher answer prediction accuracy, we propose a coarse-to-fine answer prediction approach further, which first extracts multiple answer candidates for each identified question in the coarse stage and then selects the most likely one among these candidates in the fine stage. In this way, the learning difficulty of answer prediction can be effectively reduced so that the prediction accuracy can be improved. Moreover, we introduce a spatial compatibility attention bias into the self-attention/cross-attention mechanism for \Ours{} to better model the spatial interactions between entities. With these new techniques, our proposed \Ours{} achieves state-of-the-art results on FUNSD and XFUND datasets, outperforming the previous best-performing method by 7.2\% and 13.2\% in F1 score, respectively.

Zhiyu Wu, Xiaokang Chen, Zizheng Pan et al.

We present DeepSeek-VL2, an advanced series of large Mixture-of-Experts (MoE) Vision-Language Models that significantly improves upon its predecessor, DeepSeek-VL, through two key major upgrades. For the vision component, we incorporate a dynamic tiling vision encoding strategy designed for processing high-resolution images with different aspect ratios. For the language component, we leverage DeepSeekMoE models with the Multi-head Latent Attention mechanism, which compresses Key-Value cache into latent vectors, to enable efficient inference and high throughput. Trained on an improved vision-language dataset, DeepSeek-VL2 demonstrates superior capabilities across various tasks, including but not limited to visual question answering, optical character recognition, document/table/chart understanding, and visual grounding. Our model series is composed of three variants: DeepSeek-VL2-Tiny, DeepSeek-VL2-Small and DeepSeek-VL2, with 1.0B, 2.8B and 4.5B activated parameters respectively. DeepSeek-VL2 achieves competitive or state-of-the-art performance with similar or fewer activated parameters compared to existing open-source dense and MoE-based models. Codes and pre-trained models are publicly accessible at https://github.com/deepseek-ai/DeepSeek-VL2.

Jianhong Tu, Zeyu Cui, Xiaohuan Zhou et al.

The goal of expressive Text-to-speech (TTS) is to synthesize natural speech with desired content, prosody, emotion, or timbre, in high expressiveness. Most of previous studies attempt to generate speech from given labels of styles and emotions, which over-simplifies the problem by classifying styles and emotions into a fixed number of pre-defined categories. In this paper, we introduce a new task setting, Contextual TTS (CTTS). The main idea of CTTS is that how a person speaks depends on the particular context she is in, where the context can typically be represented as text. Thus, in the CTTS task, we propose to utilize such context to guide the speech synthesis process instead of relying on explicit labels of styles and emotions. To achieve this task, we construct a synthetic dataset and develop an effective framework. Experiments show that our framework can generate high-quality expressive speech based on the given context both in synthetic datasets and real-world scenarios.

Kai Hu, Weichen Yu, Yining Li et al. · cmu

Recent research indicates that large language models (LLMs) are susceptible to jailbreaking attacks that can generate harmful content. This paper introduces a novel token-level attack method, Adaptive Dense-to-Sparse Constrained Optimization (ADC), which has been shown to successfully jailbreak multiple open-source LLMs. Drawing inspiration from the difficulties of discrete token optimization, our method relaxes the discrete jailbreak optimization into a continuous optimization process while gradually increasing the sparsity of the optimizing vectors. This technique effectively bridges the gap between discrete and continuous space optimization. Experimental results demonstrate that our method is more effective and efficient than state-of-the-art token-level methods. On Harmbench, our approach achieves the highest attack success rate on seven out of eight LLMs compared to the latest jailbreak methods. Trigger Warning: This paper contains model behavior that can be offensive in nature.

Shijun Shi, Jing Xu, Zhihang Li et al.

Recent advances in diffusion models have greatly improved pose-driven character animation. However, existing methods are limited to spatially aligned reference-pose pairs with matched skeletal structures. Handling reference-pose misalignment remains unsolved. To address this, we present One-to-All Animation, a unified framework for high-fidelity character animation and image pose transfer for references with arbitrary layouts. First, to handle spatially misaligned reference, we reformulate training as a self-supervised outpainting task that transforms diverse-layout reference into a unified occluded-input format. Second, to process partially visible reference, we design a reference extractor for comprehensive identity feature extraction. Further, we integrate hybrid reference fusion attention to handle varying resolutions and dynamic sequence lengths. Finally, from the perspective of generation quality, we introduce identity-robust pose control that decouples appearance from skeletal structure to mitigate pose overfitting, and a token replace strategy for coherent long-video generation. Extensive experiments show that our method outperforms existing approaches. The code and model are available at https://github.com/ssj9596/One-to-All-Animation.

Jiawei Wang, Kai Hu, Qiang Huo

Document structure analysis, aka document layout analysis, is crucial for understanding both the physical layout and logical structure of documents, serving information retrieval, document summarization, knowledge extraction, etc. Hierarchical Document Structure Analysis (HDSA) specifically aims to restore the hierarchical structure of documents created using authoring software with hierarchical schemas. Previous research has primarily followed two approaches: one focuses on tackling specific subtasks of HDSA in isolation, such as table detection or reading order prediction, while the other adopts a unified framework that uses multiple branches or modules, each designed to address a distinct task. In this work, we propose a unified relation prediction approach for HDSA, called UniHDSA, which treats various HDSA sub-tasks as relation prediction problems and consolidates relation prediction labels into a unified label space. This allows a single relation prediction module to handle multiple tasks simultaneously, whether at a page-level or document-level structure analysis. To validate the effectiveness of UniHDSA, we develop a multimodal end-to-end system based on Transformer architectures. Extensive experimental results demonstrate that our approach achieves state-of-the-art performance on a hierarchical document structure analysis benchmark, Comp-HRDoc, and competitive results on a large-scale document layout analysis dataset, DocLayNet, effectively illustrating the superiority of our method across all sub-tasks. The Comp-HRDoc benchmark and UniHDSA's configurations are publicly available at https://github.com/microsoft/CompHRDoc.

Kai Hu, Barnabas Poczos

In this paper, we propose a novel regularization method, RotationOut, for neural networks. Different from Dropout that handles each neuron/channel independently, RotationOut regards its input layer as an entire vector and introduces regularization by randomly rotating the vector. RotationOut can also be used in convolutional layers and recurrent layers with small modifications. We further use a noise analysis method to interpret the difference between RotationOut and Dropout in co-adaptation reduction. Using this method, we also show how to use RotationOut/Dropout together with Batch Normalization. Extensive experiments in vision and language tasks are conducted to show the effectiveness of the proposed method. Codes are available at \url{https://github.com/RotationOut/RotationOut}.

Kai Hu, Han Wang, Renhe Liu et al.

Region of Interest (ROI)-based image compression has rapidly developed due to its ability to maintain high fidelity in important regions while reducing data redundancy. However, existing compression methods primarily apply masks to suppress background information before quantization. This explicit bit allocation strategy, which uses hard gating, significantly impacts the statistical distribution of the entropy model, thereby limiting the coding performance of the compression model. In response, this work proposes an efficient ROI-based deep image compression model with implicit bit allocation. To better utilize ROI masks for implicit bit allocation, this paper proposes a novel Mask-Guided Feature Enhancement (MGFE) module, comprising a Region-Adaptive Attention (RAA) block and a Frequency-Spatial Collaborative Attention (FSCA) block. This module allows for flexible bit allocation across different regions while enhancing global and local features through frequencyspatial domain collaboration. Additionally, we use dual decoders to separately reconstruct foreground and background images, enabling the coding network to optimally balance foreground enhancement and background quality preservation in a datadriven manner. To the best of our knowledge, this is the first work to utilize implicit bit allocation for high-quality regionadaptive coding. Experiments on the COCO2017 dataset show that our implicit-based image compression method significantly outperforms explicit bit allocation approaches in rate-distortion performance, achieving optimal results while maintaining satisfactory visual quality in the reconstructed background regions.

Jiawei Wang, Kai Hu, Zhuoyao Zhong et al.

Document structure analysis (aka document layout analysis) is crucial for understanding the physical layout and logical structure of documents, with applications in information retrieval, document summarization, knowledge extraction, etc. In this paper, we concentrate on Hierarchical Document Structure Analysis (HDSA) to explore hierarchical relationships within structured documents created using authoring software employing hierarchical schemas, such as LaTeX, Microsoft Word, and HTML. To comprehensively analyze hierarchical document structures, we propose a tree construction based approach that addresses multiple subtasks concurrently, including page object detection (Detect), reading order prediction of identified objects (Order), and the construction of intended hierarchical structure (Construct). We present an effective end-to-end solution based on this framework to demonstrate its performance. To assess our approach, we develop a comprehensive benchmark called Comp-HRDoc, which evaluates the above subtasks simultaneously. Our end-to-end system achieves state-of-the-art performance on two large-scale document layout analysis datasets (PubLayNet and DocLayNet), a high-quality hierarchical document structure reconstruction dataset (HRDoc), and our Comp-HRDoc benchmark. The Comp-HRDoc benchmark will be released to facilitate further research in this field.

Ziqi Ni, Yahao Li, Kaijia Hu et al.

The rapid evolution of artificial intelligence, particularly large language models, presents unprecedented opportunities for materials science research. We proposed and developed an AI materials scientist named MatPilot, which has shown encouraging abilities in the discovery of new materials. The core strength of MatPilot is its natural language interactive human-machine collaboration, which augments the research capabilities of human scientist teams through a multi-agent system. MatPilot integrates unique cognitive abilities, extensive accumulated experience, and ongoing curiosity of human-beings with the AI agents' capabilities of advanced abstraction, complex knowledge storage and high-dimensional information processing. It could generate scientific hypotheses and experimental schemes, and employ predictive models and optimization algorithms to drive an automated experimental platform for experiments. It turns out that our system demonstrates capabilities for efficient validation, continuous learning, and iterative optimization.

Xusheng Yang, Long Zhou, Wenfu Wang et al.

We propose \textbf{U-Codec}, an \textbf{U}ltra low frame-rate neural speech \textbf{Codec} that achieves high-fidelity reconstruction and fast speech generation at an extremely low frame-rate of 5Hz (5 frames per second). Extreme compression at 5Hz typically leads to severe intelligibility and spectral detail loss, we introduce a Transformer-based inter-frame long-term dependency module and systematically explore residual vector quantization (RVQ) depth and codebook size to identify optimal configurations. Moreover, we apply U-Codec into a large language model (LLM)-based auto-regressive TTS model, which leverages global and local hierarchical architecture to effectively capture dependencies across multi-layer tokens. We extend LLM-based TTS from 3-layer RVQ at 50Hz to 32-layer RVQ at 5Hz. Experimental results demonstrate that U-Codec improves LLM-based TTS inference speed by around 3 $\times$ over high-frame-rate codecs while maintaining similarity and naturalness. These results validate the feasibility of using highly compressed 5Hz discrete tokens for fast and high-fidelity speech synthesis.

Shijun Shi, Jing Xu, Lijing Lu et al.

Existing diffusion-based video super-resolution (VSR) methods are susceptible to introducing complex degradations and noticeable artifacts into high-resolution videos due to their inherent randomness. In this paper, we propose a noise-robust real-world VSR framework by incorporating self-supervised learning and Mamba into pre-trained latent diffusion models. To ensure content consistency across adjacent frames, we enhance the diffusion model with a global spatio-temporal attention mechanism using the Video State-Space block with a 3D Selective Scan module, which reinforces coherence at an affordable computational cost. To further reduce artifacts in generated details, we introduce a self-supervised ControlNet that leverages HR features as guidance and employs contrastive learning to extract degradation-insensitive features from LR videos. Finally, a three-stage training strategy based on a mixture of HR-LR videos is proposed to stabilize VSR training. The proposed Self-supervised ControlNet with Spatio-Temporal Continuous Mamba based VSR algorithm achieves superior perceptual quality than state-of-the-arts on real-world VSR benchmark datasets, validating the effectiveness of the proposed model design and training strategies.

Jiawei Wang, Shunchi Zhang, Kai Hu et al.

Contextual Text Block Detection (CTBD) is the task of identifying coherent text blocks within the complexity of natural scenes. Previous methodologies have treated CTBD as either a visual relation extraction challenge within computer vision or as a sequence modeling problem from the perspective of natural language processing. We introduce a new framework that frames CTBD as a graph generation problem. This methodology consists of two essential procedures: identifying individual text units as graph nodes and discerning the sequential reading order relationships among these units as graph edges. Leveraging the cutting-edge capabilities of DQ-DETR for node detection, our framework innovates further by integrating a novel mechanism, a Dynamic Relation Transformer (DRFormer), dedicated to edge generation. DRFormer incorporates a dual interactive transformer decoder that deftly manages a dynamic graph structure refinement process. Through this iterative process, the model systematically enhances the graph's fidelity, ultimately resulting in improved precision in detecting contextual text blocks. Comprehensive experimental evaluations conducted on both SCUT-CTW-Context and ReCTS-Context datasets substantiate that our method achieves state-of-the-art results, underscoring the effectiveness and potential of our graph generation framework in advancing the field of CTBD.

Kai Hu, Jiawei Wang, Weihong Lin et al.

Existing methods for Visual Information Extraction (VIE) from form-like documents typically fragment the process into separate subtasks, such as key information extraction, key-value pair extraction, and choice group extraction. However, these approaches often overlook the hierarchical structure of form documents, including hierarchical key-value pairs and hierarchical choice groups. To address these limitations, we present a new perspective, reframing VIE as a relation prediction problem and unifying labels of different tasks into a single label space. This unified approach allows for the definition of various relation types and effectively tackles hierarchical relationships in form-like documents. In line with this perspective, we present UniVIE, a unified model that addresses the VIE problem comprehensively. UniVIE functions using a coarse-to-fine strategy. It initially generates tree proposals through a tree proposal network, which are subsequently refined into hierarchical trees by a relation decoder module. To enhance the relation prediction capabilities of UniVIE, we incorporate two novel tree constraints into the relation decoder: a tree attention mask and a tree level embedding. Extensive experimental evaluations on both our in-house dataset HierForms and a publicly available dataset SIBR, substantiate that our method achieves state-of-the-art results, underscoring the effectiveness and potential of our unified approach in advancing the field of VIE.

Weihong Lin, Qifang Gao, Lei Sun et al.

Recent grid-based document representations like BERTgrid allow the simultaneous encoding of the textual and layout information of a document in a 2D feature map so that state-of-the-art image segmentation and/or object detection models can be straightforwardly leveraged to extract key information from documents. However, such methods have not achieved comparable performance to state-of-the-art sequence- and graph-based methods such as LayoutLM and PICK yet. In this paper, we propose a new multi-modal backbone network by concatenating a BERTgrid to an intermediate layer of a CNN model, where the input of CNN is a document image and the BERTgrid is a grid of word embeddings, to generate a more powerful grid-based document representation, named ViBERTgrid. Unlike BERTgrid, the parameters of BERT and CNN in our multimodal backbone network are trained jointly. Our experimental results demonstrate that this joint training strategy improves significantly the representation ability of ViBERTgrid. Consequently, our ViBERTgrid-based key information extraction approach has achieved state-of-the-art performance on real-world datasets.

Kai Hu, Wentong Liao, Michael Ying Yang et al.

Text-to-image synthesis (T2I) aims to generate photo-realistic images which are semantically consistent with the text descriptions. Existing methods are usually built upon conditional generative adversarial networks (GANs) and initialize an image from noise with sentence embedding, and then refine the features with fine-grained word embedding iteratively. A close inspection of their generated images reveals a major limitation: even though the generated image holistically matches the description, individual image regions or parts of somethings are often not recognizable or consistent with words in the sentence, e.g. "a white crown". To address this problem, we propose a novel framework Semantic-Spatial Aware GAN for synthesizing images from input text. Concretely, we introduce a simple and effective Semantic-Spatial Aware block, which (1) learns semantic-adaptive transformation conditioned on text to effectively fuse text features and image features, and (2) learns a semantic mask in a weakly-supervised way that depends on the current text-image fusion process in order to guide the transformation spatially. Experiments on the challenging COCO and CUB bird datasets demonstrate the advantage of our method over the recent state-of-the-art approaches, regarding both visual fidelity and alignment with input text description.

Jian Zhu, Kai Hu, Mamoun Filali et al.

Smart contracts are the artifact of the blockchain that provide immutable and verifiable specifications of physical transactions. Solidity is a domain-specific programming language with the purpose of defining smart contracts. It aims at reducing the transaction costs occasioned by the execution of contracts on the distributed ledgers such as the Ethereum. However, Solidity contracts need to adhere safety and security requirements that require formal verification and certification. This paper proposes a method to meet such requirements by translating Solidity contracts to Event-B models, supporting certification. To that purpose, we define a restrained Solidity subset and a transfer function which translates Solidity contracts to Event-B models. Then we take advantage of Event-B method capabilities to refine models at different levels of abstraction to verify Solidity contracts' properties. And we can verify the generated proof obligations of the Event-B model with the help of the Rodin platform.

Shuo Tian, Lianhua Qu, Kai Hu et al.

Liquid State Machine (LSM), also known as the recurrent version of Spiking Neural Networks (SNN), has attracted great research interests thanks to its high computational power, biological plausibility from the brain, simple structure and low training complexity. By exploring the design space in network architectures and parameters, recent works have demonstrated great potential for improving the accuracy of LSM model with low complexity. However, these works are based on manually-defined network architectures or predefined parameters. Considering the diversity and uniqueness of brain structure, the design of LSM model should be explored in the largest search space possible. In this paper, we propose a Neural Architecture Search (NAS) based framework to explore both architecture and parameter design space for automatic dataset-oriented LSM model. To handle the exponentially-increased design space, we adopt a three-step search for LSM, including multi-liquid architecture search, variation on the number of neurons and parameters search such as percentage connectivity and excitatory neuron ratio within each liquid. Besides, we propose to use Simulated Annealing (SA) algorithm to implement the three-step heuristic search. Three datasets, including image dataset of MNIST and NMNIST and speech dataset of FSDD, are used to test the effectiveness of our proposed framework. Simulation results show that our proposed framework can produce the dataset-oriented optimal LSM models with high accuracy and low complexity. The best classification accuracy on the three datasets is 93.2%, 92.5% and 84% respectively with only 1000 spiking neurons, and the network connections can be averagely reduced by 61.4% compared with a single LSM. Moreover, we find that the total quantity of neurons in optimal LSM models on three datasets can be further reduced by 20% with only about 0.5% accuracy loss.

Kai Hu, Bhiksha Raj

Capturing spatiotemporal dynamics is an essential topic in video recognition. In this paper, we present learnable higher-order operations as a generic family of building blocks for capturing spatiotemporal dynamics from RGB input video space. Similar to higher-order functions, the weights of higher-order operations are themselves derived from the data with learnable parameters. Classical architectures such as residual learning and network-in-network are first-order operations where weights are directly learned from the data. Higher-order operations make it easier to capture context-sensitive patterns, such as motion. Self-attention models are also higher-order operations, but the attention weights are mostly computed from an affine operation or dot product. Learnable higher-order operations can be more generic and flexible. Experimentally, we show that on the task of video recognition, our higher-order models can achieve results on par with or better than the existing state-of-the-art methods on Something-Something (V1 and V2), Kinetics and Charades datasets.

Kai Hu, Zhijian Ou, Min Hu et al.

Conditional random fields (CRFs) have been shown to be one of the most successful approaches to sequence labeling. Various linear-chain neural CRFs (NCRFs) are developed to implement the non-linear node potentials in CRFs, but still keeping the linear-chain hidden structure. In this paper, we propose NCRF transducers, which consists of two RNNs, one extracting features from observations and the other capturing (theoretically infinite) long-range dependencies between labels. Different sequence labeling methods are evaluated over POS tagging, chunking and NER (English, Dutch). Experiment results show that NCRF transducers achieve consistent improvements over linear-chain NCRFs and RNN transducers across all the four tasks, and can improve state-of-the-art results.

Xiaoteng Zhang, Yixin Bao, Feiyun Zhang et al.

In this paper, we introduce our submissions for the tasks of trimmed activity recognition (Kinetics) and trimmed event recognition (Moments in Time) for Activitynet Challenge 2018. In the two tasks, non-local neural networks and temporal segment networks are implemented as our base models. Multi-modal cues such as RGB image, optical flow and acoustic signal have also been used in our method. We also propose new non-local-based models for further improvement on the recognition accuracy. The final submissions after ensembling the models achieve 83.5% top-1 accuracy and 96.8% top-5 accuracy on the Kinetics validation set, 35.81% top-1 accuracy and 62.59% top-5 accuracy on the MIT validation set.

Fen Xiao, Wenzheng Deng, Liangchan Peng et al.

Salient object detection is a fundamental problem and has been received a great deal of attentions in computer vision. Recently deep learning model became a powerful tool for image feature extraction. In this paper, we propose a multi-scale deep neural network (MSDNN) for salient object detection. The proposed model first extracts global high-level features and context information over the whole source image with recurrent convolutional neural network (RCNN). Then several stacked deconvolutional layers are adopted to get the multi-scale feature representation and obtain a series of saliency maps. Finally, we investigate a fusion convolution module (FCM) to build a final pixel level saliency map. The proposed model is extensively evaluated on four salient object detection benchmark datasets. Results show that our deep model significantly outperforms other 12 state-of-the-art approaches.