Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks94 from public data
- PolarMask: Single Shot Instance Segmentation With Polar Representation641
This paper introduces an anchor-box free and single shot instance segmentation method, which is conceptually simple, fully convolutional and can be used by easily embedding it into most off-the-shelf detection methods.
- FOTS: Fast Oriented Text Spotting with a Unified Network539
This work proposes a unified end-to-end trainable Fast Oriented Text Spotting (FOTS) network for simultaneous detection and recognition, sharing computation and visual information among the two complementary tasks, and introduces RoIRotate to share convolutional features between detection and Recognition.
- Scene Text Image Super-Resolution in the Wild186
A new Text Super-Resolution Network termed TSRN, with three novel modules is developed, which improves the recognition accuracy by over 13% of CRNN, and by nearly 9.0% of ASTER and MORAN compared to synthetic SR data.
- PAN++: Towards Efficient and Accurate End-to-End Spotting of Arbitrarily-Shaped Text177
An end-to-end text spotting framework, termed PAN++, which can efficiently detect and recognize text of arbitrary shapes in natural scenes, based on the kernel representation that reformulates a text line as a text kernel surrounded by peripheral pixels.
- Norm-Based Curriculum Learning for Neural Machine Translation135
This paper aims to improve the efficiency of training an NMT by introducing a novel norm-based curriculum learning method that uses the norm (aka length or module) of a word embedding as a measure of the difficulty of the sentence, the competence of the model, and the weight of the sentences.
- Parameter-Efficient and Student-Friendly Knowledge Distillation90
Experiments on various benchmarks including CV and NLP show that PESF-KD can significantly reduce the training cost while obtaining competitive results compared to advanced online distillation methods.
- 78
- Rejuvenating Low-Frequency Words: Making the Most of Parallel Data in Non-Autoregressive Translation55
This work directly exposes the raw data into NAT by leveraging pretraining to rejuvenate more alignments for low-frequency target words, and combines these complementary approaches as a new training strategy for further boosting NAT performance.
- Meta-Curriculum Learning for Domain Adaptation in Neural Machine Translation51
Experimental results on 10 different low-resource domains show that meta-curriculum learning can improve the translation performance of both familiar and unfamiliar domains.
- 50
- DocStruct: A Multimodal Method to Extract Hierarchy Structure in Document for General Form Understanding48
This work utilizes the state-of-the-art models and design targeted extraction modules to extract multimodal features from semantic contents, layout information, and visual images, and adopts an asymmetric algorithm and negative sampling in the model.
- Progressive Multi-Granularity Training for Non-Autoregressive Translation47
This study empirically shows that NAT models are prone to learn fine-grained lower-mode knowledge, such as words and phrases, compared with sentences, and proposes progressive multi-granularity training for NAT, which improves phrase translation accuracy and model reordering ability, resulting in better translation quality against strong NAT baselines.
- 47
- Who Wrote This? The Key to Zero-Shot LLM-Generated Text Detection Is GECScore45
This paper proposes a simple yet effective black-box zero-shot detection approach based on the observation that, from the perspective of LLMs, human-written texts typically contain more grammatical errors than LLM-generated texts.
- Can LLMs Learn Uncertainty on Their Own? Expressing Uncertainty Effectively in A Self-Training Manner41
An uncertainty-aware instruction tuning (UaIT) method is proposed, aligning LLMs’ perception with the probabilistic uncertainty of the generation to instigate a heightened awareness of self-uncertainty in LLMs, enabling them to express uncertainty more effectively.
- DelTA: An Online Document-Level Translation Agent Based on Multi-Level Memory38
Results indicate that DelTA significantly outperforms strong baselines in terms of translation consistency and quality across four open/closed-source LLMs and two representative document translation datasets, achieving an increase in consistency scores by up to 4.58 percentage points and in COMET scores by up to 3.16 points on average.
- Shared-Private Bilingual Word Embeddings for Neural Machine Translation38
Experiments demonstrate that the proposed shared-private bilingual word embeddings provide a significant performance boost over the strong baselines with dramatically fewer model parameters.
- 38
- TemplateGEC: Improving Grammatical Error Correction with Detection Template34
A novel method, TemplateGEC, is proposed, which capitalizes on the capabilities of both Seq2Edit and Sequ2Seq frameworks in error detection and correction respectively, and reveals the potential of the method in performing human-in-the-loop GEC.
- 34
- 29
- On the Complementarity between Pre-Training and Back-Translation for Neural Machine Translation29
Experimental results show that PT and BT are nicely complementary to each other, establishing state-of-the-art performances on the WMT16 English-Romanian and English-Russian benchmarks.
- AE TextSpotter: Learning Visual and Linguistic Representation for Ambiguous Text Spotting28
This work proposes a novel text spotter, named Ambiguity Eliminating Text Spotter (AE TextSpotter), which learns both visual and linguistic features to significantly reduce ambiguity in text detection, and is the first time to improve text detection by using a language model.
- ZhihuRank: A Topic-Sensitive Expert Finding Algorithm in Community Question Answering Websites28
A topic-sensitive probabilistic model to estimate the user authority ranking for each question is proposed, based on the link analysis technique and topical similarities between users and questions, which is more effective than existing approaches.
- Knowledge Editing with Dynamic Knowledge Graphs for Multi-Hop Question Answering26
KEDKG is introduced, a novel knowledge editing method that leverages a dynamic knowledge graph for MHQA, designed to ensure the reliability of answers and surpasses previous state-of-the-art models, delivering more accurate and reliable answers in environments with dynamic information.
- On the Copying Behaviors of Pre-Training for Neural Machine Translation26
Extensive experiments on both in-domain and out-of-domain benchmarks show that the copying penalty method consistently improves translation performance by controlling copying behaviors for pre-training based NMT models.
- Jailbreak-R1: Exploring the Jailbreak Capabilities of LLMs via Reinforcement Learning25
ourapproach is proposed, a novel automated red teaming training framework that utilizes reinforcement learning to explore and generate more effective attack prompts while balancing their diversity.
- Social emotion classification based on noise-aware training25
This work proposes a new architecture named PCNN, which utilizes two cascading convolutional layers to model the word-phrase relation and the phrase-sentence relation, and presents a Bayesian-based model named WMCM to learn document-level semantic features.
- 21
- 21
- 21
- TransGEC: Improving Grammatical Error Correction with Translationese20
Experimental results on the CoNLL14 and BEA19 English, NLPCC18 Chinese, Falko-MERLIN German, and RULEC-GEC Russian GEC benchmarks show that the proposed translationese approach consistently improves correction accuracy over strong baselines, and is helpful for overcoming mainstream correction difficulties.
- 17
- LRQuant: Learnable and Robust Post-Training Quantization for Large Language Models15
This paper pioneeringly introduces Test-time adaptation (TTA) into LLM quantization, which allows for rapid model adaptation during testing to improve generalization performance and finds that by us-ing the TTA method, it can achieve better results on test sets than directly using test sets for calibration in some cases.
- kNN-TL: k-Nearest-Neighbor Transfer Learning for Low-Resource Neural Machine Translation15
A k-Nearest-Neighbor Transfer Learning (kNN-TL) approach for low-resource NMT, which leverages the parent knowledge throughout the entire developing process of the child model, and which outperforms strong baselines on four low- resource translation tasks.
- 13
- 13
- Social Emotion Classification via Reader Perspective Weighted Model13
The concept of emotional entropy is proposed to estimate the weight and tackle the issue of noisy documents and the topic assignment is also used to distinguish different emotional senses of the same word.
- 12
- Revisiting Commonsense Reasoning in Machine Translation: Training, Evaluation and Challenge12
This paper confirms the effectiveness of incorporating pretrained knowledge into NMT models and subsequently utilizing these models as robust testbeds for investigating CR in NMT and proposes a novel entity-aware evaluation method that takes into account both the NMT candidate and important entities in the candidate, which is more aligned with human judgement.
- Can LMs Generalize to Future Data? An Empirical Analysis on Text Summarization11
This work proposes TempoSum, a novel benchmark that contains data samples from 2010 to 2022, to understand the temporal generalization ability of abstractive summarization models and shows that parametric knowledge stored in summarization Models significantly affects the faithfulness of the generated summaries on future data.
- BLISS: Robust Sequence-to-Sequence Learning via Self-Supervised Input Representation11
This work proposes a framework-level robust sequence-to-sequence learning approach, named BLISS, via self-supervised input representation, which has the great potential to complement the data-level augmentation approaches.
- 11
- 10
- 9
- Test-time Adaptation for Machine Translation Evaluation by Uncertainty Minimization8
This paper aims to address the inference bias of neural metrics through uncertainty minimization during test time, without requiring additional data, by employing the prediction uncertainty of the current data as a signal to update a small fraction of parameters during testTime.
- Difficulty-Aware Machine Translation Evaluation8
A novel difficulty-aware MT evaluation metric is proposed, expanding the evaluation dimension by taking translation difficulty into consideration, and shows that the proposed method outperforms commonly used MT metrics in terms of human correlation.
- Dynamic Sampling that Adapts: Self-Aware Iterative Data Persistent Optimization for Mathematical Reasoning7
SAI-DPO (Self-Aware Iterative Data Persistent Optimization), a dynamic sampling framework that aligns training data with the model's intrinsic competence, achieves state-of-the-art efficiency with significantly less data.
- Towards Demonstration-Aware Large Language Models for Machine Translation7
A novel approach that integrates demonstration-aware training and inference strategies within the framework of tuning-based LTMs, hereby referred to as demonstration-aware LTMs, which mitigates the negative impacts traditionally associated with demonstrations and secures substantial improvements in translation accuracy, particularly in domain-specific and document-level translation tasks.
- 7
- 7
- REA-RL: Reflection-Aware Online Reinforcement Learning for Efficient Reasoning6
This work proposes REA-RL, which introduces a small reflection model for efficient scaling in online training, offering both parallel sampling and sequential revision, and a reflection reward is designed to further prevent LRMs from favoring short yet non-reflective responses.
- Think Dense, Not Long: Dynamic Decoupled Conditional Advantage for Efficient Reasoning5
Dynamic Decoupled Conditional Advantage (DDCA) computes length advantages conditionally within the correct-response cluster to eliminate baseline dilution, and dynamically scales the penalty strength using the group pass rate as a proxy for difficulty.
- 5
- Speech Sense Disambiguation: Tackling Homophone Ambiguity in End-to-End Speech Translation5
AmbigST is introduced, an innovative homophone-aware contrastive learning approach that integrates a homophone-aware masking strategy and achieves SOTA results on BLEU scores for English to German, Spanish, and French ST tasks, underlining its effectiveness in reducing speech sense ambiguity.
- Curriculum Consistency Learning for Conditional Sentence Generation5
This work introduces Curriculum Consistency Learning (CCL), a novel strategy that guides models to learn consistency in alignment with their current capacity to differentiate between features, and indicates that models utilizing CCL are particularly adept at managing complex instances.
- 5
- 5
- The SYSU System for CCPR 2016 Multimodal Emotion Recognition Challenge5
A residual network architecture within the convolutional neural networks (CNN) framework to improve the facial expression recognition performance is proposed and the openSMILE based utterance level feature and MFCC GMM based zero-order statistics feature are extracted for the subsequent SVM modeling in the speech based subsystem.
- 4
- 4
- Clustering Pseudo Language Family in Multilingual Translation Models with Fisher Information Matrix4
An innovative method that leverages the fisher information matrix (FIM) to cluster language families, anchored on the multilingual translation model's characteristics, and finds that employing these pseudo language families enhances performance over conventional language families in adapting a multilingualtranslation model to unfamiliar language pairs.
- A Context Awareness Hierarchical Attention Network for Next POI Recommendation in IoT Environment4
A Context Geographical-Temporal-Social Awareness Hierarchical Attention Network (CGTS-HAN) model is proposed and the experimental results on Foursquare and Yelp real datasets show that the AUC, precision and recall of CGTS- HAN are better than the comparison models, which proves the effectiveness and superiority of the model.
- 3
- Domain-Aware k-Nearest-Neighbor Knowledge Distillation for Machine Translation3
A novel domain-aware k NN-KD method, which filters out domain-relevant neighborhood knowledge for learning in the distillation process, and exclusively utilizes the neighborhood knowledge of the original model, eliminating the need for establishing any additional datas-tores.
- 3
- Mitigating Context-Memory Conflicts in LLMs Through Dynamic Cognitive Reconciliation Decoding2
This paper proposes a two-stage decoding method called Dynamic Cognitive Reconciliation Decoding (DCRD), to predict and mitigate context-memory conflicts, and demonstrates that DCRD outperforms all baselines, achieving state-of-the-art performance.
- Noise Improves Multimodal Machine Translation: Rethinking the Role of Visual Context2
This work revisits commonly used image-must and image-free MMT approaches and proposes a novel visual type regularization method and introduces two probing tasks to analyze whether and how visual features influence a strong MMT model.
- 2
- 2
- Loong: A Human-Like Long Document Translation Agent with Observe-and-Act Adaptive Context Selection1
A human-like long document translation agent called Loong, which leverages a 3E memory module (Essence-Exemplar-Entity) to store summaries, sentence pairs, and entity records as historical context and performs deep reasoning to adaptively identify the optimal context for translation guidance.
- Exposing the Cracks: Vulnerabilities of Retrieval-Augmented LLM-based Machine Translation1
A noise synthesis framework and new metrics are proposed to systematically evaluate REAL-MT’s reliability across high-, medium-, and low-resource language pairs, highlighting the limitations of current approaches and underscoring the need for self-verifying integration mechanisms.
- Orchestrating Prompt Expertise: Enhancing Knowledge Distillation via Expert-Guided Tuning1
MoE-KD is presented, a simple but effective framework that produces supervision for training the student model from one single teacher model, which fixes the above problems and improves effectiveness and extended MoE KD to lifelong learning scenarios, proposing a lightweight solution for catastrophic forgetting.
- 1
- 1
- 1
- Context-aware Dynamic Block.1
A dynamic inference method called Context-aware Dynamic Block (CDB), which provides more path selection choices in terms of network width and depth during inference, which can be easily incorporated into most modern network architectures.
- User authority ranking models for community question answering1
This study proposes three probabilistic models to rank the user authority of each question and uses a real-world dataset from Zhihu, a popular community question answering website in China to conduct experiments, showing that the model outperforms other baseline methods in ranking the user authority.
- –
- –
- –
- –
- Domain Adaptive Machine Translation with Synthetic Feedback for Large Language Models–
A pipeline that collects in-domain translations from LLMs and generates synthetic, human-like feedback for revising these translations is proposed that demonstrates the effectiveness of the pipeline in tailoring in-domain translations and improving translation performance compared to direct translation instructions.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-15. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.