Search Results for author: Deng Cai

Found 155 papers, 78 papers with code

Retrieval is Accurate Generation

no code implementations27 Feb 2024 Bowen Cao, Deng Cai, Leyang Cui, Xuxin Cheng, Wei Bi, Yuexian Zou, Shuming Shi

To address this, we propose to initialize the training oracles using linguistic heuristics and, more importantly, bootstrap the oracles through iterative self-reinforcement.

Language Modelling Retrieval +1

NeRF-Det++: Incorporating Semantic Cues and Perspective-aware Depth Supervision for Indoor Multi-View 3D Detection

1 code implementation22 Feb 2024 Chenxi Huang, Yuenan Hou, Weicai Ye, Di Huang, Xiaoshui Huang, Binbin Lin, Deng Cai, Wanli Ouyang

We project the freely available 3D segmentation annotations onto the 2D plane and leverage the corresponding 2D semantic maps as the supervision signal, significantly enhancing the semantic awareness of multi-view detectors.

Depth Estimation Depth Prediction +1

Model Compression and Efficient Inference for Large Language Models: A Survey

no code implementations15 Feb 2024 Wenxiao Wang, Wei Chen, Yicong Luo, Yongliu Long, Zhengkai Lin, Liye Zhang, Binbin Lin, Deng Cai, Xiaofei He

However, Large language models have two prominent characteristics compared to smaller models: (1) Most of compression algorithms require finetuning or even retraining the model after compression.

Knowledge Distillation Model Compression +1

A Thorough Examination of Decoding Methods in the Era of LLMs

no code implementations10 Feb 2024 Chufan Shi, Haoran Yang, Deng Cai, Zhisong Zhang, Yifan Wang, Yujiu Yang, Wai Lam

Decoding methods play an indispensable role in converting language models from next-token predictors into practical task solvers.


LiFi: Lightweight Controlled Text Generation with Fine-Grained Control Codes

no code implementations10 Feb 2024 Chufan Shi, Deng Cai, Yujiu Yang

In the rapidly evolving field of text generation, the demand for more precise control mechanisms has become increasingly apparent.

Attribute Language Modelling +1

UniHDA: Towards Universal Hybrid Domain Adaptation of Image Generators

no code implementations23 Jan 2024 Hengjia Li, Yang Liu, Yuqi Lin, Zhanwei Zhang, Yibo Zhao, weihang Pan, Tu Zheng, Zheng Yang, Yuchun Jiang, Boxi Wu, Deng Cai

In this paper, we propose UniHDA, a unified and versatile framework for generative hybrid domain adaptation with multi-modal references from multiple domains.

Attribute Domain Adaptation

Knowledge Fusion of Large Language Models

1 code implementation19 Jan 2024 Fanqi Wan, Xinting Huang, Deng Cai, Xiaojun Quan, Wei Bi, Shuming Shi

In this paper, we introduce the notion of knowledge fusion for LLMs, aimed at combining the capabilities of existing LLMs and transferring them into a single LLM.

Code Generation

Inferflow: an Efficient and Highly Configurable Inference Engine for Large Language Models

1 code implementation16 Jan 2024 Shuming Shi, Enbo Zhao, Deng Cai, Leyang Cui, Xinting Huang, Huayang Li

We present Inferflow, an efficient and highly configurable inference engine for large language models (LLMs).


Reasons to Reject? Aligning Language Models with Judgments

1 code implementation22 Dec 2023 Weiwen Xu, Deng Cai, Zhisong Zhang, Wai Lam, Shuming Shi

As humans, we consistently engage in interactions with our peers and receive feedback in the form of natural language.

Regulating Intermediate 3D Features for Vision-Centric Autonomous Driving

1 code implementation19 Dec 2023 Junkai Xu, Liang Peng, Haoran Cheng, Linxuan Xia, Qi Zhou, Dan Deng, Wei Qian, Wenxiao Wang, Deng Cai

To resolve this problem, we propose to regulate intermediate dense 3D features with the help of volume rendering.

Autonomous Driving

GPT4Video: A Unified Multimodal Large Language Model for lnstruction-Followed Understanding and Safety-Aware Generation

no code implementations25 Nov 2023 Zhanyu Wang, Longyue Wang, Zhen Zhao, Minghao Wu, Chenyang Lyu, Huayang Li, Deng Cai, Luping Zhou, Shuming Shi, Zhaopeng Tu

While the recent advances in Multimodal Large Language Models (MLLMs) constitute a significant leap forward in the field, these models are predominantly confined to the realm of input-side multimodal comprehension, lacking the capacity for multimodal content generation.

Instruction Following Language Modelling +7

StrategyLLM: Large Language Models as Strategy Generators, Executors, Optimizers, and Evaluators for Problem Solving

no code implementations15 Nov 2023 Chang Gao, Haiyun Jiang, Deng Cai, Shuming Shi, Wai Lam

Most existing chain-of-thought (CoT) prompting methods suffer from the issues of generalizability and consistency, as they often rely on instance-specific solutions that may not be applicable to other cases and lack task-level consistency in their reasoning steps.


Specialist or Generalist? Instruction Tuning for Specific NLP Tasks

no code implementations23 Oct 2023 Chufan Shi, Yixuan Su, Cheng Yang, Yujiu Yang, Deng Cai

Although instruction tuning has proven to be a data-efficient method for transforming LLMs into such generalist models, their performance still lags behind specialist models trained exclusively for specific tasks.


Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models

1 code implementation3 Sep 2023 Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong Chen, Longyue Wang, Anh Tuan Luu, Wei Bi, Freda Shi, Shuming Shi

While large language models (LLMs) have demonstrated remarkable capabilities across a range of downstream tasks, a significant concern revolves around their propensity to exhibit hallucinations: LLMs occasionally generate content that diverges from the user input, contradicts previously generated context, or misaligns with established world knowledge.

Hallucination World Knowledge

MonoNeRD: NeRF-like Representations for Monocular 3D Object Detection

1 code implementation ICCV 2023 Junkai Xu, Liang Peng, Haoran Cheng, Hao Li, Wei Qian, Ke Li, Wenxiao Wang, Deng Cai

To the best of our knowledge, this work is the first to introduce volume rendering for M3D, and demonstrates the potential of implicit reconstruction for image-based 3D perception.

Monocular 3D Object Detection Object +1

TeCH: Text-guided Reconstruction of Lifelike Clothed Humans

1 code implementation16 Aug 2023 Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Jiaxiang Tang, Deng Cai, Justus Thies

But how to effectively capture all visual attributes of an individual from a single image, which are sufficient to reconstruct unseen areas (e. g., the back view)?

Descriptive Question Answering +1

NormKD: Normalized Logits for Knowledge Distillation

1 code implementation1 Aug 2023 Zhihao Chi, Tu Zheng, Hengjia Li, Zheng Yang, Boxi Wu, Binbin Lin, Deng Cai

In this paper, we restudy the hyper-parameter temperature and figure out its incapability to distill the knowledge from each sample sufficiently when it is a single value.

Image Classification Knowledge Distillation

Disco-Bench: A Discourse-Aware Evaluation Benchmark for Language Modelling

no code implementations16 Jul 2023 Longyue Wang, Zefeng Du, Donghuai Liu, Deng Cai, Dian Yu, Haiyun Jiang, Yan Wang, Leyang Cui, Shuming Shi, Zhaopeng Tu

Modeling discourse -- the linguistic phenomena that go beyond individual sentences, is a fundamental yet challenging aspect of natural language processing (NLP).

Language Modelling Sentence

Copy Is All You Need

1 code implementation13 Jul 2023 Tian Lan, Deng Cai, Yan Wang, Heyan Huang, Xian-Ling Mao

The dominant text generation models compose the output by sequentially selecting words from a fixed vocabulary.

Domain Adaptation Language Modelling +1

PandaGPT: One Model To Instruction-Follow Them All

1 code implementation25 May 2023 Yixuan Su, Tian Lan, Huayang Li, Jialu Xu, Yan Wang, Deng Cai

To do so, PandaGPT combines the multimodal encoders from ImageBind and the large language models from Vicuna.

Instruction Following

A Frustratingly Simple Decoding Method for Neural Text Generation

1 code implementation22 May 2023 Haoran Yang, Deng Cai, Huayang Li, Wei Bi, Wai Lam, Shuming Shi

We introduce a frustratingly simple, super efficient and surprisingly effective decoding method, which we call Frustratingly Simple Decoding (FSD), for neural text generation.

Language Modelling Text Generation

Multi-Task Instruction Tuning of LLaMa for Specific Scenarios: A Preliminary Study on Writing Assistance

no code implementations22 May 2023 Yue Zhang, Leyang Cui, Deng Cai, Xinting Huang, Tao Fang, Wei Bi

Proprietary Large Language Models (LLMs), such as ChatGPT, have garnered significant attention due to their exceptional capabilities in handling a diverse range of tasks.

Instruction Following

Neural Collapse Inspired Federated Learning with Non-iid Data

no code implementations27 Mar 2023 Chenxi Huang, Liang Xie, Yibo Yang, Wenxiao Wang, Binbin Lin, Deng Cai

One of the challenges in federated learning is the non-independent and identically distributed (non-iid) characteristics between heterogeneous devices, which cause significant differences in local updates and affect the performance of the central server.

Federated Learning

OBMO: One Bounding Box Multiple Objects for Monocular 3D Object Detection

1 code implementation20 Dec 2022 Chenxi Huang, Tong He, Haidong Ren, Wenxiao Wang, Binbin Lin, Deng Cai

Unfortunately, the network cannot accurately distinguish different depths from such non-discriminative visual features, resulting in unstable depth training.

Monocular 3D Object Detection object-detection

Self-supervised and Weakly Supervised Contrastive Learning for Frame-wise Action Representations

no code implementations6 Dec 2022 Minghao Chen, Renbo Tu, Chenxi Huang, Yuqi Lin, Boxi Wu, Deng Cai

In this paper, we introduce a new framework of contrastive action representation learning (CARL) to learn frame-wise action representation in a self-supervised or weakly-supervised manner, especially for long videos.

Action Classification Contrastive Learning +4

Boosting Semi-Supervised 3D Object Detection with Semi-Sampling

no code implementations14 Nov 2022 Xiaopei Wu, Yang Zhao, Liang Peng, Hua Chen, Xiaoshui Huang, Binbin Lin, Haifeng Liu, Deng Cai, Wanli Ouyang

When training a teacher-student semi-supervised framework, we randomly select gt samples and pseudo samples to both labeled frames and unlabeled frames, making a strong data augmentation for them.

3D Object Detection Data Augmentation +2

$N$-gram Is Back: Residual Learning of Neural Text Generation with $n$-gram Language Model

1 code implementation26 Oct 2022 Huayang Li, Deng Cai, Jin Xu, Taro Watanabe

The combination of $n$-gram and neural LMs not only allows the neural part to focus on the deeper understanding of language but also provides a flexible way to customize an LM by switching the underlying $n$-gram model without changing the neural model.

Domain Adaptation Language Modelling +2

Retrofitting Multilingual Sentence Embeddings with Abstract Meaning Representation

1 code implementation18 Oct 2022 Deng Cai, Xin Li, Jackie Chun-Sing Ho, Lidong Bing, Wai Lam

Unlike most prior work that only evaluates the ability to measure semantic similarity, we present a thorough evaluation of existing multilingual sentence embeddings and our improved versions, which include a collection of five transfer tasks in different downstream applications.

Semantic Similarity Semantic Textual Similarity +2

Towards Efficient Adversarial Training on Vision Transformers

no code implementations21 Jul 2022 Boxi Wu, Jindong Gu, Zhifeng Li, Deng Cai, Xiaofei He, Wei Liu

Vision Transformer (ViT), as a powerful alternative to Convolutional Neural Network (CNN), has received much attention.

DID-M3D: Decoupling Instance Depth for Monocular 3D Object Detection

1 code implementation18 Jul 2022 Liang Peng, Xiaopei Wu, Zheng Yang, Haifeng Liu, Deng Cai

Therefore, we propose to reformulate the instance depth to the combination of the instance visual surface depth (visual depth) and the instance attribute depth (attribute depth).

Attribute Data Augmentation +4

Automatic Prosody Annotation with Pre-Trained Text-Speech Model

1 code implementation16 Jun 2022 Ziqian Dai, Jianwei Yu, Yan Wang, Nuo Chen, Yanyao Bian, Guangzhi Li, Deng Cai, Dong Yu

Prosodic boundary plays an important role in text-to-speech synthesis (TTS) in terms of naturalness and readability.

Speech Synthesis Text-To-Speech Synthesis

Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generation

2 code implementations6 Jun 2022 Jin Xu, Xiaojiang Liu, Jianhao Yan, Deng Cai, Huayang Li, Jian Li

While large-scale neural language models, such as GPT2 and BART, have achieved impressive results on various text generation tasks, they tend to get stuck in undesirable sentence-level loops with maximization-based decoding algorithms (\textit{e. g.}, greedy search).

Sentence Text Generation +1

Frame-wise Action Representations for Long Videos via Sequence Contrastive Learning

1 code implementation CVPR 2022 Minghao Chen, Fangyun Wei, Chong Li, Deng Cai

In this paper, we introduce a novel contrastive action representation learning (CARL) framework to learn frame-wise action representations, especially for long videos, in a self-supervised manner.

Action Classification Contrastive Learning +4

Linearizing Transformer with Key-Value Memory

no code implementations23 Mar 2022 Yizhe Zhang, Deng Cai

We demonstrate that MemSizer provides an improved balance between efficiency and accuracy over the vanilla transformer and other efficient transformer variants in three typical sequence generation tasks, including machine translation, abstractive text summarization, and language modeling.

Abstractive Text Summarization Language Modelling +2

CLRNet: Cross Layer Refinement Network for Lane Detection

3 code implementations CVPR 2022 Tu Zheng, Yifei HUANG, Yang Liu, Wenjian Tang, Zheng Yang, Deng Cai, Xiaofei He

In this way, we can exploit more contextual information to detect lanes while leveraging local detailed lane features to improve localization accuracy.

Lane Detection

Sparse Fuse Dense: Towards High Quality 3D Detection with Depth Completion

1 code implementation CVPR 2022 Xiaopei Wu, Liang Peng, Honghui Yang, Liang Xie, Chenxi Huang, Chengqi Deng, Haifeng Liu, Deng Cai

Many multi-modal methods are proposed to alleviate this issue, while different representations of images and point clouds make it difficult to fuse them, resulting in suboptimal performance.

3D Object Detection Data Augmentation +3

WeakM3D: Towards Weakly Supervised Monocular 3D Object Detection

1 code implementation ICLR 2022 Liang Peng, Senbo Yan, Boxi Wu, Zheng Yang, Xiaofei He, Deng Cai

This network is learned by minimizing our newly-proposed 3D alignment loss between the 3D box estimates and the corresponding RoI LiDAR points.

Monocular 3D Object Detection Object +2

A Survey on Retrieval-Augmented Text Generation

no code implementations2 Feb 2022 Huayang Li, Yixuan Su, Deng Cai, Yan Wang, Lemao Liu

Recently, retrieval-augmented text generation attracted increasing attention of the computational linguistics community.

Machine Translation Response Generation +3

TopNet: Learning from Neural Topic Model to Generate Long Stories

no code implementations14 Dec 2021 Yazheng Yang, Boyuan Pan, Deng Cai, Huan Sun

In particular, instead of directly generating a story, we first learn to map the short text input to a low-dimensional topic distribution (which is pre-assigned by a topic model).

Story Generation

Exploring Dense Retrieval for Dialogue Response Selection

1 code implementation13 Oct 2021 Tian Lan, Deng Cai, Yan Wang, Yixuan Su, Heyan Huang, Xian-Ling Mao

In this study, we present a solution to directly select proper responses from a large corpus or even a nonparallel corpus that only consists of unpaired sentences, using a dense retrieval model.

Conversational Response Selection Retrieval

Multilingual AMR Parsing with Noisy Knowledge Distillation

1 code implementation Findings (EMNLP) 2021 Deng Cai, Xin Li, Jackie Chun-Sing Ho, Lidong Bing, Wai Lam

We study multilingual AMR parsing from the perspective of knowledge distillation, where the aim is to learn and improve a multilingual AMR parser by using an existing English parser as its teacher.

AMR Parsing Knowledge Distillation

Digging Into Output Representation for Monocular 3D Object Detection

no code implementations29 Sep 2021 Liang Peng, Senbo Yan, Chenxi Huang, Xiaofei He, Deng Cai

This characteristic indicates that monocular 3D detection is inherently different from other typical detection tasks that have the same dimensional input and output.

Monocular 3D Object Detection Object +1

Exploiting Reasoning Chains for Multi-hop Science Question Answering

1 code implementation Findings (EMNLP) 2021 Weiwen Xu, Yang Deng, Huihui Zhang, Deng Cai, Wai Lam

We propose a novel Chain Guided Retriever-reader ({\tt CGR}) framework to model the reasoning chain for multi-hop Science Question Answering.

Science Question Answering

ASR-GLUE: A New Multi-task Benchmark for ASR-Robust Natural Language Understanding

no code implementations30 Aug 2021 Lingyun Feng, Jianwei Yu, Deng Cai, Songxiang Liu, Haitao Zheng, Yan Wang

%To facilitate the research on ASR-robust general language understanding, In this paper, we propose ASR-GLUE benchmark, a new collection of 6 different NLU tasks for evaluating the performance of models under ASR error across 3 different levels of background noise and 6 speakers with various voice characteristics.

Automatic Speech Recognition Automatic Speech Recognition (ASR) +3

CrossFormer: A Versatile Vision Transformer Hinging on Cross-scale Attention

3 code implementations ICLR 2022 Wenxiao Wang, Lu Yao, Long Chen, Binbin Lin, Deng Cai, Xiaofei He, Wei Liu

On the one hand, CEL blends each embedding with multiple patches of different scales, providing the self-attention module itself with cross-scale features.

Image Classification Instance Segmentation +4

Learning to Affiliate: Mutual Centralized Learning for Few-shot Classification

1 code implementation CVPR 2022 Yang Liu, Weifeng Zhang, Chao Xiang, Tu Zheng, Deng Cai, Xiaofei He

Few-shot learning (FSL) aims to learn a classifier that can be easily adapted to accommodate new tasks not seen during training, given only a few examples.

Classification Few-Shot Learning

Attacking Adversarial Attacks as A Defense

no code implementations9 Jun 2021 Boxi Wu, Heng Pan, Li Shen, Jindong Gu, Shuai Zhao, Zhifeng Li, Deng Cai, Xiaofei He, Wei Liu

In this work, we find that the adversarial attacks can also be vulnerable to small perturbations.

Salient Object Ranking with Position-Preserved Attention

1 code implementation ICCV 2021 Hao Fang, Daoxin Zhang, Yi Zhang, Minghao Chen, Jiawei Li, Yao Hu, Deng Cai, Xiaofei He

In this paper, we study the Salient Object Ranking (SOR) task, which manages to assign a ranking order of each detected object according to its visual saliency.

Image Cropping Instance Segmentation +7

Dynamic Semantic Graph Construction and Reasoning for Explainable Multi-hop Science Question Answering

1 code implementation Findings (ACL) 2021 Weiwen Xu, Huihui Zhang, Deng Cai, Wai Lam

Our framework contains three new ideas: (a) {\tt AMR-SG}, an AMR-based Semantic Graph, constructed by candidate fact AMRs to uncover any hop relations among question, answer and multiple facts.

graph construction Knowledge Graphs +4

Assessing Dialogue Systems with Distribution Distances

1 code implementation Findings (ACL) 2021 Jiannan Xiang, Yahui Liu, Deng Cai, Huayang Li, Defu Lian, Lemao Liu

An important aspect of developing dialogue systems is how to evaluate and compare the performance of different systems.

Dialogue Evaluation

Discriminative-Generative Dual Memory Video Anomaly Detection

no code implementations29 Apr 2021 Xin Guo, Zhongming Jin, Chong Chen, Helei Nie, Jianqiang Huang, Deng Cai, Xiaofei He, Xiansheng Hua

In this paper, we propose a DiscRiminative-gEnerative duAl Memory (DREAM) anomaly detection model to take advantage of a few anomalies and solve data imbalance.

Anomaly Detection Video Anomaly Detection

Lidar Point Cloud Guided Monocular 3D Object Detection

1 code implementation19 Apr 2021 Liang Peng, Fei Liu, Zhengxu Yu, Senbo Yan, Dan Deng, Zheng Yang, Haifeng Liu, Deng Cai

We delve into this underlying mechanism and then empirically find that: concerning the label accuracy, the 3D location part in the label is preferred compared to other parts of labels.

Monocular 3D Object Detection Object +1

OCM3D: Object-Centric Monocular 3D Object Detection

no code implementations13 Apr 2021 Liang Peng, Fei Liu, Senbo Yan, Xiaofei He, Deng Cai

Image-only and pseudo-LiDAR representations are commonly used for monocular 3D object detection.

Monocular 3D Object Detection Object +1

SCALoss: Side and Corner Aligned Loss for Bounding Box Regression

1 code implementation1 Apr 2021 Tu Zheng, Shuai Zhao, Yang Liu, Zili Liu, Deng Cai

In this paper, we propose Side Overlap~(SO) loss by maximizing the side overlap of two bounding boxes, which puts more penalty for low overlapping bounding box cases.

object-detection Object Detection +1

X-view: Non-egocentric Multi-View 3D Object Detector

no code implementations24 Mar 2021 Liang Xie, Guodong Xu, Deng Cai, Xiaofei He

3D object detection algorithms for autonomous driving reason about 3D obstacles either from 3D birds-eye view or perspective view or both.

3D Object Detection Autonomous Driving +3

DMN4: Few-shot Learning via Discriminative Mutual Nearest Neighbor Neural Network

no code implementations15 Mar 2021 Yang Liu, Tu Zheng, Jie Song, Deng Cai, Xiaofei He

In this paper, we argue that a Mutual Nearest Neighbor (MNN) relation should be established to explicitly select the query descriptors that are most relevant to each task and discard less relevant ones from aggregative clutters in FSL.

Few-Shot Learning

ES-Net: Erasing Salient Parts to Learn More in Re-Identification

no code implementations10 Mar 2021 Dong Shen, Shuai Zhao, Jinming Hu, Hao Feng, Deng Cai, Xiaofei He

In this paper, we propose a novel network, Erasing-Salient Net (ES-Net), to learn comprehensive features by erasing the salient areas in an image.

Complementary Pseudo Labels For Unsupervised Domain Adaptation On Person Re-identification

no code implementations29 Jan 2021 Hao Feng, Minghao Chen, Jinming Hu, Dong Shen, Haifeng Liu, Deng Cai

In this paper, to complement these low recall neighbor pseudo labels, we propose a joint learning framework to learn better feature embeddings via high precision neighbor pseudo labels and high recall group pseudo labels.

Person Re-Identification Unsupervised Domain Adaptation

Dialogue Response Selection with Hierarchical Curriculum Learning

1 code implementation ACL 2021 Yixuan Su, Deng Cai, Qingyu Zhou, Zibo Lin, Simon Baker, Yunbo Cao, Shuming Shi, Nigel Collier, Yan Wang

As for IC, it progressively strengthens the model's ability in identifying the mismatching information between the dialogue context and a response candidate.

Conversational Response Selection

Narrative Incoherence Detection

no code implementations21 Dec 2020 Deng Cai, Yizhe Zhang, Yichen Huang, Wai Lam, Bill Dolan

We propose the task of narrative incoherence detection as a new arena for inter-sentential semantic understanding: Given a multi-sentence narrative, decide whether there exist any semantic discrepancies in the narrative flow.

Sentence Sentence Embedding

EasyTransfer -- A Simple and Scalable Deep Transfer Learning Platform for NLP Applications

2 code implementations18 Nov 2020 Minghui Qiu, Peng Li, Chengyu Wang, Hanjie Pan, Ang Wang, Cen Chen, Xianyan Jia, Yaliang Li, Jun Huang, Deng Cai, Wei Lin

The literature has witnessed the success of leveraging Pre-trained Language Models (PLMs) and Transfer Learning (TL) algorithms to a wide range of Natural Language Processing (NLP) applications, yet it is not easy to build an easy-to-use and scalable TL toolkit for this purpose.

Compiler Optimization Conversational Question Answering +1

Reducing the Teacher-Student Gap via Spherical Knowledge Disitllation

1 code implementation15 Oct 2020 Jia Guo, Minghao Chen, Yao Hu, Chen Zhu, Xiaofei He, Deng Cai

We investigate this problem by study the gap of confidence between teacher and student.

Knowledge Distillation

Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework

no code implementations10 Oct 2020 Wenxiao Wang, Minghao Chen, Shuai Zhao, Long Chen, Jinming Hu, Haifeng Liu, Deng Cai, Xiaofei He, Wei Liu

Specifically, it first casts the relationships between a certain model's accuracy and depth/width/resolution into a polynomial regression and then maximizes the polynomial to acquire the optimal values for the three dimensions.

Network Pruning Neural Architecture Search +1

Do Wider Neural Networks Really Help Adversarial Robustness?

1 code implementation NeurIPS 2021 Boxi Wu, Jinghui Chen, Deng Cai, Xiaofei He, Quanquan Gu

Previous empirical results suggest that adversarial training requires wider networks for better performances.

Adversarial Robustness

Interpretable Real-Time Win Prediction for Honor of Kings, a Popular Mobile MOBA Esport

no code implementations14 Aug 2020 Zelong Yang, Zhufeng Pan, Yan Wang, Deng Cai, Xiaojiang Liu, Shuming Shi, Shao-Lun Huang

With the rapid prevalence and explosive development of MOBA esports (Multiplayer Online Battle Arena electronic sports), much research effort has been devoted to automatically predicting game results (win predictions).


Apparel-invariant Feature Learning for Apparel-changed Person Re-identification

no code implementations14 Aug 2020 Zhengxu Yu, Yilun Zhao, Bin Hong, Zhongming Jin, Jianqiang Huang, Deng Cai, Xiaofei He, Xian-Sheng Hua

Therefore, it is critical to learn an apparel-invariant person representation under cases like cloth changing or several persons wearing similar clothes.

Person Re-Identification Representation Learning

Learning to Caricature via Semantic Shape Transform

1 code implementation12 Aug 2020 Wenqing Chu, Wei-Chih Hung, Yi-Hsuan Tsai, Yu-Ting Chang, Yijun Li, Deng Cai, Ming-Hsuan Yang

Caricature is an artistic drawing created to abstract or exaggerate facial features of a person.


Describe What to Change: A Text-guided Unsupervised Image-to-Image Translation Approach

1 code implementation10 Aug 2020 Yahui Liu, Marco De Nadai, Deng Cai, Huayang Li, Xavier Alameda-Pineda, Nicu Sebe, Bruno Lepri

Our proposed model disentangles the image content from the visual attributes, and it learns to modify the latter using the textual description, before generating a new image from the content and the modified attribute representation.

Attribute Image Captioning +3

Out-of-distribution Generalization via Partial Feature Decorrelation

no code implementations30 Jul 2020 Xin Guo, Zhengxu Yu, Chao Xiang, Zhongming Jin, Jianqiang Huang, Deng Cai, Xiaofei He, Xian-Sheng Hua

Most deep-learning-based image classification methods assume that all samples are generated under an independent and identically distributed (IID) setting.

Classification General Classification +3

Adversarial Mutual Information for Text Generation

1 code implementation ICML 2020 Boyuan Pan, Yazheng Yang, Kaizhao Liang, Bhavya Kailkhura, Zhongming Jin, Xian-Sheng Hua, Deng Cai, Bo Li

Recent advances in maximizing mutual information (MI) between the source and target have demonstrated its effectiveness in text generation.

Text Generation

AMR Parsing via Graph-Sequence Iterative Inference

3 code implementations ACL 2020 Deng Cai, Wai Lam

We propose a new end-to-end model that treats AMR parsing as a series of dual decisions on the input sequence and the incrementally constructed graph.

AMR Parsing Language Modelling

The World is Not Binary: Learning to Rank with Grayscale Data for Dialogue Response Selection

no code implementations EMNLP 2020 Zibo Lin, Deng Cai, Yan Wang, Xiaojiang Liu, Hai-Tao Zheng, Shuming Shi

Despite that response selection is naturally a learning-to-rank problem, most prior works take a point-wise view and train binary classifiers for this task: each response candidate is labeled either relevant (one) or irrelevant (zero).

Conversational Response Selection Learning-To-Rank +2

Stylistic Dialogue Generation via Information-Guided Reinforcement Learning Strategy

no code implementations5 Apr 2020 Yixuan Su, Deng Cai, Yan Wang, Simon Baker, Anna Korhonen, Nigel Collier, Xiaojiang Liu

To enable better balance between the content quality and the style, we introduce a new training strategy, know as Information-Guided Reinforcement Learning (IG-RL).

Dialogue Generation reinforcement-learning +2

Prototype-to-Style: Dialogue Generation with Style-Aware Editing on Retrieval Memory

no code implementations5 Apr 2020 Yixuan Su, Yan Wang, Simon Baker, Deng Cai, Xiaojiang Liu, Anna Korhonen, Nigel Collier

A stylistic response generator then takes the prototype and the desired language style as model input to obtain a high-quality and stylistic response.

Dialogue Generation Information Retrieval +1

Bi-Decoder Augmented Network for Neural Machine Translation

no code implementations14 Jan 2020 Boyuan Pan, Yazheng Yang, Zhou Zhao, Yueting Zhuang, Deng Cai

Neural Machine Translation (NMT) has become a popular technology in recent years, and the encoder-decoder framework is the mainstream among all the methods.

Machine Translation NMT +1

Adversarial-Learned Loss for Domain Adaptation

1 code implementation4 Jan 2020 Minghao Chen, Shuai Zhao, Haifeng Liu, Deng Cai

In order to combine the strengths of these two methods, we propose a novel method called Adversarial-Learned Loss for Domain Adaptation (ALDA).

Domain Adaptation Pseudo Label

DBP: Discrimination Based Block-Level Pruning for Deep Model Acceleration

no code implementations21 Dec 2019 Wenxiao Wang, Shuai Zhao, Minghao Chen, Jinming Hu, Deng Cai, Haifeng Liu

The dominant pruning methods, filter-level pruning methods, evaluate their performance through the reduction ratio of computations and deem that a higher reduction ratio of computations is equivalent to a higher acceleration ratio in terms of inference time.

Network Pruning

Graph Transformer for Graph-to-Sequence Learning

1 code implementation18 Nov 2019 Deng Cai, Wai Lam

The dominant graph-to-sequence transduction models employ graph neural networks for graph representation learning, where the structural information is reflected by the receptive field of neurons.

AMR-to-Text Generation Graph Representation Learning +4

PI-RCNN: An Efficient Multi-sensor 3D Object Detector with Point-based Attentive Cont-conv Fusion Module

no code implementations14 Nov 2019 Liang Xie, Chao Xiang, Zhengxu Yu, Guodong Xu, Zheng Yang, Deng Cai, Xiaofei He

Moreover, based on the PACF module, we propose a 3D multi-sensor multi-task network called Pointcloud-Image RCNN(PI-RCNN as brief), which handles the image segmentation and 3D object detection tasks.

3D Object Detection Image Segmentation +4

Retrieval-guided Dialogue Response Generation via a Matching-to-Generation Framework

no code implementations IJCNLP 2019 Deng Cai, Yan Wang, Wei Bi, Zhaopeng Tu, Xiaojiang Liu, Shuming Shi

End-to-end sequence generation is a popular technique for developing open domain dialogue systems, though they suffer from the \textit{safe response problem}.

Response Generation Retrieval

Region Mutual Information Loss for Semantic Segmentation

2 code implementations NeurIPS 2019 Shuai Zhao, Yang Wang, Zheng Yang, Deng Cai

In this paper, we develop a region mutual information (RMI) loss to model the dependencies among pixels more simply and efficiently.

Semantic Segmentation

Correlation Maximized Structural Similarity Loss for Semantic Segmentation

no code implementations19 Oct 2019 Shuai Zhao, Boxi Wu, Wenqing Chu, Yao Hu, Deng Cai

Inspired by the widely-used structural similarity (SSIM) index in image quality assessment, we use the linear correlation between two images to quantify their structural similarity.

Generative Adversarial Network Image Quality Assessment +2

Domain Adaptation for Semantic Segmentation with Maximum Squares Loss

1 code implementation ICCV 2019 Minghao Chen, Hongyang Xue, Deng Cai

However, when applying the entropy minimization to UDA for semantic segmentation, the gradient of the entropy is biased towards samples that are easy to transfer.

Semantic Segmentation Unsupervised Domain Adaptation

Core Semantic First: A Top-down Approach for AMR Parsing

1 code implementation IJCNLP 2019 Deng Cai, Wai Lam

The output graph spans the nodes by the distance to the root, following the intuition of first grasping the main ideas then digging into more details.

AMR Parsing Sentence

Training-Time-Friendly Network for Real-Time Object Detection

6 code implementations2 Sep 2019 Zili Liu, Tu Zheng, Guodong Xu, Zheng Yang, Haifeng Liu, Deng Cai

Experiments on MS COCO show that our TTFNet has great advantages in balancing training time, inference speed, and accuracy.

Object object-detection +1

Charge-Based Prison Term Prediction with Deep Gating Network

no code implementations IJCNLP 2019 Huajie Chen, Deng Cai, Wei Dai, Zehui Dai, Yadong Ding

Judgment prediction for legal cases has attracted much research efforts for its practice use, of which the ultimate goal is prison term prediction.

feature selection

Neural Machine Translation with Noisy Lexical Constraints

no code implementations13 Aug 2019 Huayang Li, Guoping Huang, Deng Cai, Lemao Liu

Experiments show that our approach can indeed improve the translation quality with the automatically generated constraints.

Machine Translation Open-Ended Question Answering +1

Progressive Transfer Learning

1 code implementation7 Aug 2019 Zhengxu Yu, Dong Shen, Zhongming Jin, Jianqiang Huang, Deng Cai, Xian-Sheng Hua

Model fine-tuning is a widely used transfer learning approach in person Re-identification (ReID) applications, which fine-tuning a pre-trained feature extraction model into the target scenario instead of training a model from scratch.

Image Classification Person Re-Identification +1

Reinforced Dynamic Reasoning for Conversational Question Generation

1 code implementation ACL 2019 Boyuan Pan, Hao Li, Ziyu Yao, Deng Cai, Huan Sun

This paper investigates a new task named Conversational Question Generation (CQG) which is to generate a question based on a passage and a conversation history (i. e., previous turns of question-answer pairs).

Question Answering Question Generation +2

Improving Semantic Segmentation via Dilated Affinity

no code implementations16 Jul 2019 Boxi Wu, Shuai Zhao, Wenqing Chu, Zheng Yang, Deng Cai

To be specific, our method explicitly requires the network to predict semantic segmentation as well as dilated affinity, which is a sparse version of pair-wise pixel affinity.

Segmentation Semantic Segmentation

High Dimensional Similarity Search with Satellite System Graph: Efficiency, Scalability, and Unindexed Query Compatibility

2 code implementations13 Jul 2019 Cong Fu, Changxu Wang, Deng Cai

However, we find there are several limitations with NSG: 1) NSG has no theoretical guarantee on nearest neighbor search when the query is not indexed in the database; 2) NSG is too sparse which harms the search performance.

Information Retrieval Retrieval

A Minimax Game for Instance based Selective Transfer Learning

no code implementations1 Jul 2019 Bo wang, Minghui Qiu, Xisen Wang, Yaliang Li, Yu Gong, Xiaoyi Zeng, Jung Huang, Bo Zheng, Deng Cai, Jingren Zhou

To the best of our knowledge, this is the first to build a minimax game based model for selective transfer learning.

Retrieval Text Retrieval +1

Localizing Unseen Activities in Video via Image Query

no code implementations28 Jun 2019 Zhu Zhang, Zhou Zhao, Zhijie Lin, Jingkuan Song, Deng Cai

Thus, we consider a new task to localize unseen activities in videos via image queries, named Image-Based Activity Localization.

Action Localization Video Understanding

COP: Customized Deep Model Compression via Regularized Correlation-Based Filter-Level Pruning

1 code implementation25 Jun 2019 Wenxiao Wang, Cong Fu, Jishun Guo, Deng Cai, Xiaofei He

2) Cross-layer filter comparison is unachievable since the importance is defined locally within each layer.

Neural Network Compression

Query-based Interactive Recommendation by Meta-Path and Adapted Attention-GRU

1 code implementation24 Jun 2019 Yu Zhu, Yu Gong, Qingwen Liu, Yingcai Ma, Wenwu Ou, Junxiong Zhu, Beidou Wang, Ziyu Guan, Deng Cai

A novel query-based interactive recommender system is proposed in this paper, where \textbf{personalized questions are accurately generated from millions of automatically constructed questions} in Step 1, and \textbf{the recommendation is ensured to be closely-related to users' feedback} in Step 2.

Recommendation Systems Retrieval

Weakly-supervised Caricature Face Parsing through Domain Adaptation

1 code implementation13 May 2019 Wenqing Chu, Wei-Chih Hung, Yi-Hsuan Tsai, Deng Cai, Ming-Hsuan Yang

However, current state-of-the-art face parsing methods require large amounts of labeled data on the pixel-level and such process for caricature is tedious and labor-intensive.

Attribute Caricature +3

Chinese Word Segmentation: Another Decade Review (2007-2017)

no code implementations18 Jan 2019 Hai Zhao, Deng Cai, Changning Huang, Chunyu Kit

This paper reviews the development of Chinese word segmentation (CWS) in the most recent decade, 2007-2017.

Chinese Word Segmentation

Translating a Math Word Problem to an Expression Tree

1 code implementation14 Nov 2018 Lei Wang, Yan Wang, Deng Cai, Dongxiang Zhang, Xiaojiang Liu

Moreover, we analyze the performance of three popular SEQ2SEQ models on the math word problem solving.

Math Math Word Problem Solving

Textually Guided Ranking Network for Attentional Image Retweet Modeling

no code implementations24 Oct 2018 Zhou Zhao, Hanbing Zhan, Lingtao Meng, Jun Xiao, Jun Yu, Min Yang, Fei Wu, Deng Cai

In this paper, we study the problem of image retweet prediction in social media, which predicts the image sharing behavior that the user reposts the image tweets from their followees.

Skeleton-to-Response: Dialogue Generation Guided by Retrieval Memory

1 code implementation NAACL 2019 Deng Cai, Yan Wang, Victoria Bi, Zhaopeng Tu, Xiaojiang Liu, Wai Lam, Shuming Shi

Such models rely on insufficient information for generating a specific response since a certain query could be answered in multiple ways.

Dialogue Generation Information Retrieval +3

Language Style Transfer from Sentences with Arbitrary Unknown Styles

no code implementations13 Aug 2018 Yanpeng Zhao, Wei Bi, Deng Cai, Xiaojiang Liu, Kewei Tu, Shuming Shi

Then, by recombining the content with the target style, we decode a sentence aligned in the target domain.

Sentence Sentence ReWriting +1

A Brand-level Ranking System with the Customized Attention-GRU Model

no code implementations23 May 2018 Yu Zhu, Junxiong Zhu, Jie Hou, Yongliang Li, Beidou Wang, Ziyu Guan, Deng Cai

In e-commerce websites like Taobao, brand is playing a more important role in influencing users' decision of click/purchase, partly because users are now attaching more importance to the quality of products and brand is an indicator of quality.

Feature Engineering Test

Addressing the Item Cold-start Problem by Attribute-driven Active Learning

no code implementations23 May 2018 Yu Zhu, Jinhao Lin, Shibi He, Beidou Wang, Ziyu Guan, Haifeng Liu, Deng Cai

Both content information (e. g. item attributes) and initial user ratings are valuable for seizing users' preferences on a new item.

Active Learning Attribute +2

PixelLink: Detecting Scene Text via Instance Segmentation

5 code implementations4 Jan 2018 Dan Deng, Haifeng Liu, Xuelong. Li, Deng Cai

Most state-of-the-art scene text detection algorithms are deep learning based methods that depend on bounding box regression and perform at least two kinds of predictions: text/non-text classification and location regression.

Instance Segmentation regression +5

On the Diversity of Realistic Image Synthesis

1 code implementation20 Dec 2017 Zichen Yang, Haifeng Liu, Deng Cai

Experimental results show that images synthesized by our approach are significantly more diverse than that of the current existing works and equipping our diversity loss does not degrade the reality of the base networks.

Colorization Image Generation +1

A Revisit on Deep Hashings for Large-scale Content Based Image Retrieval

no code implementations16 Nov 2017 Deng Cai, Xiuye Gu, Chaoqi Wang

However, there are serious flaws in the evaluations of existing deep hashing papers: (1) The datasets they used are too small and simple to simulate the real CBIR situation.

Content-Based Image Retrieval Deep Hashing

Dialogue Act Recognition via CRF-Attentive Structured Network

no code implementations SIGIR 2018 Zheqian Chen, Rongqin Yang, Zhou Zhao, Deng Cai, Xiaofei He

Dialogue Act Recognition (DAR) is a challenging problem in dialogue interpretation, which aims to attach semantic labels to utterances and characterize the speaker's intention.

Dialogue Act Classification Dialogue Interpretation +1

Keyword-based Query Comprehending via Multiple Optimized-Demand Augmentation

no code implementations1 Nov 2017 Boyuan Pan, Hao Li, Zhou Zhao, Deng Cai, Xiaofei He

In this paper, we propose a novel neural network system that consists a Demand Optimization Model based on a passage-attention neural machine translation and a Reader Model that can find the answer given the optimized question.

Machine Reading Comprehension Machine Translation +2

Smarnet: Teaching Machines to Read and Comprehend Like Human

no code implementations8 Oct 2017 Zheqian Chen, Rongqin Yang, Bin Cao, Zhou Zhao, Deng Cai, Xiaofei He

Machine Comprehension (MC) is a challenging task in Natural Language Processing field, which aims to guide the machine to comprehend a passage and answer the given question.

Question Answering Reading Comprehension +1

Learning Graph-Level Representation for Drug Discovery

2 code implementations12 Sep 2017 Junying Li, Deng Cai, Xiaofei He

Molecules can be represented as an undirected graph, and we can utilize graph convolution networks to predication molecular properties.

Drug Discovery General Classification +1

Fast Approximate Nearest Neighbor Search With The Navigating Spreading-out Graph

2 code implementations1 Jul 2017 Cong Fu, Chao Xiang, Changxu Wang, Deng Cai

In this paper, to further improve the search-efficiency and scalability of graph-based methods, we start by introducing four aspects: (1) ensuring the connectivity of the graph; (2) lowering the average out-degree of the graph for fast traversal; (3) shortening the search path; and (4) reducing the index size.

Deep Rotation Equivariant Network

2 code implementations24 May 2017 Junying Li, Zichen Yang, Haifeng Liu, Deng Cai

Recently, learning equivariant representations has attracted considerable research attention.

Rotated MNIST

The Forgettable-Watcher Model for Video Question Answering

no code implementations3 May 2017 Hongyang Xue, Zhou Zhao, Deng Cai

Then we propose a TGIF-QA dataset for video question answering with the help of automatic question generation.

Question Answering Question Generation +3

Fast and Accurate Neural Word Segmentation for Chinese

1 code implementation ACL 2017 Deng Cai, Hai Zhao, Zhisong Zhang, Yuan Xin, Yongjian Wu, Feiyue Huang

Neural models with minimal feature engineering have achieved competitive performance against traditional methods for the task of Chinese word segmentation.

Chinese Word Segmentation Feature Engineering +1

A Revisit of Hashing Algorithms for Approximate Nearest Neighbor Search

2 code implementations22 Dec 2016 Deng Cai

However, many existing hashing papers only report the performance with the code length shorter than 128.

Question Retrieval for Community-based Question Answering via Heterogeneous Network Integration Learning

no code implementations24 Nov 2016 Zheqian Chen, Chi Zhang, Zhou Zhao, Deng Cai

The challenges in this task are the lexical gaps between questions for the word ambiguity and word mismatch problem.

Question Answering Retrieval

Relational Multi-Manifold Co-Clustering

no code implementations16 Nov 2016 Ping Li, Jiajun Bu, Chun Chen, Zhanying He, Deng Cai

In this study, we focus on improving the co-clustering performance via manifold ensemble learning, which is able to maximally approximate the intrinsic manifolds of both the sample and feature spaces.

Clustering Ensemble Learning

Constrained Low-Rank Learning Using Least Squares-Based Regularization

no code implementations15 Nov 2016 Ping Li, Jun Yu, Meng Wang, Luming Zhang, Deng Cai, Xuelong. Li

To achieve this goal, we cast the problem into a constrained rank minimization framework by adopting the least squares regularization.

General Classification Image Categorization +2

EFANNA : An Extremely Fast Approximate Nearest Neighbor Search Algorithm Based on kNN Graph

5 code implementations23 Sep 2016 Cong Fu, Deng Cai

In this paper, we propose EFANNA, an extremely fast approximate nearest neighbor search algorithm based on $k$NN Graph.

graph construction

Scaling Up Sparse Support Vector Machines by Simultaneous Feature and Sample Reduction

1 code implementation ICML 2017 Weizhong Zhang, Bin Hong, Wei Liu, Jieping Ye, Deng Cai, Xiaofei He, Jie Wang

By noting that sparse SVMs induce sparsities in both feature and sample spaces, we propose a novel approach, which is based on accurate estimations of the primal and dual optima of sparse SVMs, to simultaneously identify the inactive features and samples that are guaranteed to be irrelevant to the outputs.

Neural Word Segmentation Learning for Chinese

1 code implementation ACL 2016 Deng Cai, Hai Zhao

Most previous approaches to Chinese word segmentation formalize this problem as a character-based sequence labeling task where only contextual information within fixed sized local windows and simple interactions between adjacent tags can be captured.

Chinese Word Segmentation Feature Engineering +1

Depth Image Inpainting: Improving Low Rank Matrix Completion with Low Gradient Regularization

1 code implementation20 Apr 2016 Hongyang Xue, Shengming Zhang, Deng Cai

The proposed low gradient regularization is integrated with the low rank regularization into the low rank low gradient approach for depth image inpainting.

Image Inpainting Low-Rank Matrix Completion

Deep Feature Based Contextual Model for Object Detection

no code implementations14 Apr 2016 Wenqing Chu, Deng Cai

Object detection is one of the most active areas in computer vision, which has made significant improvement in recent years.

Object object-detection +1

Compressed Hashing

no code implementations CVPR 2013 Yue Lin, Rong Jin, Deng Cai, Shuicheng Yan, Xuelong. Li

Recent studies have shown that hashing methods are effective for high dimensional nearest neighbor search.

Cannot find the paper you are looking for? You can Submit a new open access paper.