Liang Huang

He Bai, Renjie Zheng, Junkun Chen et al. · apple-ml

Recently, speech representation learning has improved many speech-related tasks such as speech recognition, speech classification, and speech-to-text translation. However, all the above tasks are in the direction of speech understanding, but for the inverse direction, speech synthesis, the potential of representation learning is yet to be realized, due to the challenging nature of generating high-quality speech. To address this problem, we propose our framework, Alignment-Aware Acoustic-Text Pretraining (A$^3$T), which reconstructs masked acoustic signals with text input and acoustic-text alignment during training. In this way, the pretrained model can generate high quality reconstructed spectrogram, which can be applied to the speech editing and unseen speaker TTS directly. Experiments show A$^3$T outperforms SOTA models on speech editing, and improves multi-speaker speech synthesis without the external speaker verification model.

Xiaoran Fan, Chao Pang, Tian Yuan et al. · apple-ml

Speech representation learning has improved both speech understanding and speech synthesis tasks for single language. However, its ability in cross-lingual scenarios has not been explored. In this paper, we extend the pretraining method for cross-lingual multi-speaker speech synthesis tasks, including cross-lingual multi-speaker voice cloning and cross-lingual multi-speaker speech editing. We propose a speech-text joint pretraining framework, where we randomly mask the spectrogram and the phonemes given a speech example and its transcription. By learning to reconstruct the masked parts of the input in different languages, our model shows great improvements over speaker-embedding-based multi-speaker TTS methods. Moreover, our framework is end-to-end for both the training and the inference without any finetuning effort. In cross-lingual multi-speaker voice cloning and cross-lingual multi-speaker speech editing tasks, our experiments show that our model outperforms speaker-embedding-based multi-speaker TTS methods.

Weijian Chen, Haotian Liu, Yangshen Deng et al.

On-disk graph-based approximate nearest neighbor search (ANNS) is essential for large-scale, high-dimensional vector retrieval, yet its performance is widely recognized to be limited by the prohibitive I/O costs. Interestingly, we observed that the performance of on-disk graph-based index systems is compute-bound, not I/O-bound, with the rising of the vector data dimensionality (e.g., hundreds or thousands). This insight uncovers a significant optimization opportunity: existing on-disk graph-based index systems universally target I/O reduction and largely overlook computational overhead, which leaves a substantial performance improvement space. In this work, we propose AlayaLaser, an efficient on-disk graph-based index system for large-scale high-dimensional vector similarity search. In particular, we first conduct performance analysis on existing on-disk graph-based index systems via the adapted roofline model, then we devise a novel on-disk data layout in AlayaLaser to effectively alleviate the compute-bound, which is revealed by the above roofline model analysis, by exploiting SIMD instructions on modern CPUs. We next design a suite of optimization techniques (e.g., degree-based node cache, cluster-based entry point selection, and early dispatch strategy) to further improve the performance of AlayaLaser. We last conduct extensive experimental studies on a wide range of large-scale high-dimensional vector datasets to verify the superiority of AlayaLaser. Specifically, AlayaLaser not only surpasses existing on-disk graph-based index systems but also matches or even exceeds the performance of in-memory index systems.

Aichen Cai, Anmeng Zhang, Anyu Li et al.

We introduce JoyAI-LLM Flash, an efficient Mixture-of-Experts (MoE) language model designed to redefine the trade-off between strong performance and token efficiency in the sub-50B parameter regime. JoyAI-LLM Flash is pretrained on a massive corpus of 20 trillion tokens and further optimized through a rigorous post-training pipeline, including supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and large-scale reinforcement learning (RL) across diverse environments. To improve token efficiency, JoyAI-LLM Flash strategically balances \emph{thinking} and \emph{non-thinking} cognitive modes and introduces FiberPO, a novel RL algorithm inspired by fibration theory that decomposes trust-region maintenance into global and local components, providing unified multi-scale stability control for LLM policy optimization. To enhance architectural sparsity, the model comprises 48B total parameters while activating only 2.7B parameters per forward pass, achieving a substantially higher sparsity ratio than contemporary industry leading models of comparable scale. To further improve inference throughput, we adopt a joint training-inference co-design that incorporates dense Multi-Token Prediction (MTP) and Quantization-Aware Training (QAT). We release the checkpoints for both JoyAI-LLM-48B-A3B Base and its post-trained variants on Hugging Face to support the open-source community.

Tianshuo Zhou, David H. Mathews, Liang Huang

Motivation: RNA design aims to find RNA sequences that fold into a given target secondary structure, a problem also known as RNA inverse folding. However, not all target structures are designable. Recent advances in RNA designability have focused primarily on minimum free energy (MFE)-based criteria, while ensemble-based notions of designability remain largely underexplored. To address this gap, we introduce a theory of ensemble approximation and a probability decomposition framework for bounding the folding probabilities of RNA structures in an explainable way. We further develop a linear-time dynamic programming algorithm that efficiently searches over exponentially many decompositions and identifies the optimal one that yields the tightest probabilistic bound for a given structure. Results: Applying our methods to both native and artificial RNA structures in the ArchiveII and Eterna100 benchmarks, we obtained probability bounds that are much tighter than prior approaches. In addition, our methods further provide anatomical tools for analyzing RNA structures and understanding the sources of design difficulty at the motif level. Availability: Source code and data are available at https://github.com/shanry/RNA-Undesign. Supplementary information: Supplementary text and data are available in a separate PDF.

Guangxu Xun, Mingbo Ma, Yuchen Bian et al.

In simultaneous translation (SimulMT), the most widely used strategy is the wait-k policy thanks to its simplicity and effectiveness in balancing translation quality and latency. However, wait-k suffers from two major limitations: (a) it is a fixed policy that can not adaptively adjust latency given context, and (b) its training is much slower than full-sentence translation. To alleviate these issues, we propose a novel and efficient training scheme for adaptive SimulMT by augmenting the training corpus with adaptive prefix-to-prefix pairs, while the training complexity remains the same as that of training full-sentence translation models. Experiments on two language pairs show that our method outperforms all strong baselines in terms of translation quality and latency.

Liang Huang, Senjie Liang, Feiyang Ye et al.

Intent detection and slot filling are two main tasks in natural language understanding and play an essential role in task-oriented dialogue systems. The joint learning of both tasks can improve inference accuracy and is popular in recent works. However, most joint models ignore the inference latency and cannot meet the need to deploy dialogue systems at the edge. In this paper, we propose a Fast Attention Network (FAN) for joint intent detection and slot filling tasks, guaranteeing both accuracy and latency. Specifically, we introduce a clean and parameter-refined attention module to enhance the information exchange between intent and slot, improving semantic accuracy by more than 2%. FAN can be implemented on different encoders and delivers more accurate models at every speed level. Our experiments on the Jetson Nano platform show that FAN inferences fifteen utterances per second with a small accuracy drop, showing its effectiveness and efficiency on edge devices.

Zetian Wu, Bowen Xie, Wuyang Meng et al.

The field of sign language translation has witnessed significant progress in the translation between sign and spoken languages, but the translation between sign languages remains largely unexplored and out of reach. The latter can help 1.5 billion deaf and hard-of-hearing (DHH) people worldwide communicate across language barriers without relying on hearing interpreters or written-language fluency. The cascade approach composing separate sign-to-text, text-to-text, and text-to-sign systems suffers from error propagation and extra latency as well as the loss of information unique in the visual modality. We aim to develop direct sign-to-sign translation. However, a large-scale open-domain parallel corpus has not been curated between sign languages. To enable direct translation between sign language utterances, we use back-translation to produce synthetic sign-sign pairs from unaligned individual language utterance-sign corpora. Using this data, we jointly train a single MBART-based model for both text->sign (T2S) and sign->sign (S2S). On synthetically generated paired sets between American Sign Language (ASL), Chinese Sign Language (CSL), and German Sign Language (DGS), our direct S2S method outperforms the cascaded baseline on geometric sign error metrics (20% lower DTW-aligned MPJPE) and language matching metrics after predicted sign utterances are translated back to sentences (50% high BLEU-4) while achieving a roughly 2.3* speedup. On a small set of pre-existing cross-lingual sign data, we find similar improvements for our proposed method.

Milan Gautam, Ning Dai, Tianshuo Zhou et al.

RNA design, the task of finding a sequence that folds into a target secondary structure, has broad biological and biomedical impact but remains computationally challenging due to the exponentially large sequence space and exponentially many competing folds. Traditional approaches treat it as an optimization problem, relying on per-instance heuristics or constraint-based search. We instead reframe RNA design as conditional sequence generation and introduce a reusable neural approximator, instantiated as an autoregressive language model (LM), that maps target structures directly to sequences. We first train our model in a supervised setting on random-induced structure-sequence pairs, and then use reinforcement learning (RL) to optimize end-to-end metrics. We also propose methods to select a small subset for RL that greatly improves RL efficiency and quality. Across four datasets, our approach outperforms state-of-the-art systems on key metrics such as Boltzmann probability while being 1.7x faster, establishing conditional LM generation as a scalable, task-agnostic alternative to per-instance optimization for RNA design. Our code and data are available at https://github.com/KuNyaa/RNA-Design-LM.

Shaojie Wang, Jinghui Wang, Yinghan Cui et al.

In agentic LLM scenarios, an agent's interaction process during a single rollout often exhibits branching behaviors. Due to memory retrieval and concurrent tool executions at certain decision points, the token trajectory of one task evolves into a tree-like structure rather than a linear sequence. However, current training pipelines decompose such tree-structured trajectories into separate linear segments, treating each branch as an independent sequence. As a result, shared prefixes across these branches are repeatedly recomputed during both forward and backward passes. To address this inefficiency, we propose Tree Training, a paradigm that computes each shared prefix only once and reuses its intermediate results across related branches during both forward and backward passes, substantially improving computation efficiency in large-scale agentic training. This is achieved via (i) Tree Packing, which efficiently reuses shared computations across trajectories, and (ii) Gradient Restoration, which ensures correct gradient propagation across reused prefixes. Experiments on multiple open-source models demonstrate up to 3.9x reduction in total training time, enabling more efficient agentic LLM SFT and RL training.

Guoxin Wang, Minyu Gao, Shuai Yang et al.

Large language models (LLMs), particularly those with reasoning capabilities, have rapidly advanced in recent years, demonstrating significant potential across a wide range of applications. However, their deployment in healthcare, especially in disease reasoning tasks, is hindered by the challenge of acquiring expert-level cognitive data. In this paper, we introduce Citrus, a medical language model that bridges the gap between clinical expertise and AI reasoning by emulating the cognitive processes of medical experts. The model is trained on a large corpus of simulated expert disease reasoning data, synthesized using a novel approach that accurately captures the decision-making pathways of clinicians. This approach enables Citrus to better simulate the complex reasoning processes involved in diagnosing and treating medical conditions. To further address the lack of publicly available datasets for medical reasoning tasks, we release the last-stage training data, including a custom-built medical diagnostic dialogue dataset. This open-source contribution aims to support further research and development in the field. Evaluations using authoritative benchmarks such as MedQA, covering tasks in medical reasoning and language understanding, show that Citrus achieves superior performance compared to other models of similar size. These results highlight Citrus potential to significantly enhance medical decision support systems, providing a more accurate and efficient tool for clinical decision-making.

Wei Yu Tang, Ning Dai, Tianshuo Zhou et al.

The task of RNA design given a target structure aims to find a sequence that can fold into that structure. It is a computationally hard problem where some version(s) have been proven to be NP-hard. As a result, heuristic methods such as local search have been popular for this task, but by only exploring a fixed number of candidates. They can not keep up with the exponential growth of the design space, and often perform poorly on longer and harder-to-design structures. We instead formulate these discrete problems as continuous optimization, which starts with a distribution over all possible candidate sequences, and uses gradient descent to improve the expectation of an objective function. We define novel distributions based on coupled variables to rule out invalid sequences given the target structure and to model the correlation between nucleotides. To make it universally applicable to any objective function, we use sampling to approximate the expected objective function, to estimate the gradient, and to select the final candidate. Compared to the state-of-the-art methods, our work consistently outperforms them in key metrics such as Boltzmann probability, ensemble defect, and energy gap, especially on long and hard-to-design puzzles in the Eterna100 benchmark. Our code is available at: http://github.com/weiyutang1010/ncrna_design.

Ning Dai, Zheng Wu, Renjie Zheng et al.

Reinforcement learning (RL) with unit test feedback has enhanced large language models' (LLMs) code generation, but relies on sparse rewards provided only after complete code evaluation, limiting learning efficiency and incremental improvements. When generated code fails all unit tests, no learning signal is received, hindering progress on complex tasks. To address this, we propose a Process Reward Model (PRM) that delivers dense, line-level feedback on code correctness during generation, mimicking human code refinement and providing immediate guidance. We explore various strategies for training PRMs and integrating them into the RL framework, finding that using PRMs both as dense rewards and for value function initialization significantly boosts performance. Our experimental results also highlight the effectiveness of PRMs in enhancing RL-driven code generation, especially for long-horizon scenarios.

Ning Dai, Wei Yu Tang, Tianshuo Zhou et al.

The tasks of designing RNAs are discrete optimization problems, and several versions of these problems are NP-hard. As an alternative to commonly used local search methods, we formulate these problems as continuous optimization and develop a general framework for this optimization based on a generalization of classical partition function which we call "expected partition function". The basic idea is to start with a distribution over all possible candidate sequences, and extend the objective function from a sequence to a distribution. We then use gradient descent-based optimization methods to improve the extended objective function, and the distribution will gradually shrink towards a one-hot sequence (i.e., a single sequence). As a case study, we consider the important problem of mRNA design with wide applications in vaccines and therapeutics. While the recent work of LinearDesign can efficiently optimize mRNAs for minimum free energy (MFE), optimizing for ensemble free energy is much harder and likely intractable. Our approach can consistently improve over the LinearDesign solution in terms of ensemble free energy, with bigger improvements on longer sequences.

Zetian Wu, Tianshuo Zhou, Stefan Lee et al.

Sign language translation from text to video plays a crucial role in enabling effective communication for Deaf and hard--of--hearing individuals. A major challenge lies in generating accurate and natural body poses and movements that faithfully convey intended meanings. Prior methods often neglect the anatomical constraints and coordination patterns of human skeletal motion, resulting in rigid or biomechanically implausible outputs. To address this, we propose a novel approach that explicitly models the relationships among skeletal joints--including shoulders, arms, and hands--by incorporating geometric constraints on joint positions, bone lengths, and movement dynamics. During training, we introduce a parent-relative reweighting mechanism to enhance finger flexibility and reduce motion stiffness. Additionally, bone-pose losses and bone-length constraints enforce anatomically consistent structures. Our method narrows the performance gap between the previous best and the ground-truth oracle by 56.51%, and further reduces discrepancies in bone length and movement variance by 18.76% and 5.48%, respectively, demonstrating significant gains in anatomical realism and motion naturalness.

Xiaoyi Zhou, Liang Huang, Tong Ye et al.

This paper investigates an unmanned aerial vehicle (UAV)-assisted wireless powered mobile-edge computing (MEC) system, where the UAV powers the mobile terminals by wireless power transfer (WPT) and provides computation service for them. We aim to maximize the computation rate of terminals while ensuring fairness among them. Considering the random trajectories of mobile terminals, we propose a soft actor-critic (SAC)-based UAV trajectory planning and resource allocation (SAC-TR) algorithm, which combines off-policy and maximum entropy reinforcement learning to promote the convergence of the algorithm. We design the reward as a heterogeneous function of computation rate, fairness, and reaching of destination. Simulation results show that SAC-TR can quickly adapt to varying network environments and outperform representative benchmarks in a variety of situations.

Jiahong Yuan, Xingyu Cai, Renjie Zheng et al.

We propose a method for emotion recognition through emotiondependent speech recognition using Wav2vec 2.0. Our method achieved a significant improvement over most previously reported results on IEMOCAP, a benchmark emotion dataset. Different types of phonetic units are employed and compared in terms of accuracy and robustness of emotion recognition within and across datasets and languages. Models of phonemes, broad phonetic classes, and syllables all significantly outperform the utterance model, demonstrating that phonetic units are helpful and should be incorporated in speech emotion recognition. The best performance is from using broad phonetic classes. Further research is needed to investigate the optimal set of broad phonetic classes for the task of emotion recognition. Finally, we found that Wav2vec 2.0 can be fine-tuned to recognize coarser-grained or larger phonetic units than phonemes, such as broad phonetic classes and syllables.

Jiahong Yuan, Xingyu Cai, Dongji Gao et al.

Much of the recent literature on automatic speech recognition (ASR) is taking an end-to-end approach. Unlike English where the writing system is closely related to sound, Chinese characters (Hanzi) represent meaning, not sound. We propose factoring audio -> Hanzi into two sub-tasks: (1) audio -> Pinyin and (2) Pinyin -> Hanzi, where Pinyin is a system of phonetic transcription of standard Chinese. Factoring the audio -> Hanzi task in this way achieves 3.9% CER (character error rate) on the Aishell-1 corpus, the best result reported on this dataset so far.

Shuangli Li, Jingbo Zhou, Tong Xu et al.

Drug discovery often relies on the successful prediction of protein-ligand binding affinity. Recent advances have shown great promise in applying graph neural networks (GNNs) for better affinity prediction by learning the representations of protein-ligand complexes. However, existing solutions usually treat protein-ligand complexes as topological graph data, thus the biomolecular structural information is not fully utilized. The essential long-range interactions among atoms are also neglected in GNN models. To this end, we propose a structure-aware interactive graph neural network (SIGN) which consists of two components: polar-inspired graph attention layers (PGAL) and pairwise interactive pooling (PiPool). Specifically, PGAL iteratively performs the node-edge aggregation process to update embeddings of nodes and edges while preserving the distance and angle information among atoms. Then, PiPool is adopted to gather interactive edges with a subsequent reconstruction loss to reflect the global interactions. Exhaustive experimental study on two benchmarks verifies the superiority of SIGN.

Junkun Chen, Mingbo Ma, Renjie Zheng et al.

Simultaneous speech-to-text translation is widely useful in many scenarios. The conventional cascaded approach uses a pipeline of streaming ASR followed by simultaneous MT, but suffers from error propagation and extra latency. To alleviate these issues, recent efforts attempt to directly translate the source speech into target text simultaneously, but this is much harder due to the combination of two separate tasks. We instead propose a new paradigm with the advantages of both cascaded and end-to-end approaches. The key idea is to use two separate, but synchronized, decoders on streaming ASR and direct speech-to-text translation (ST), respectively, and the intermediate results of ASR guide the decoding policy of (but is not fed as input to) ST. During training time, we use multitask learning to jointly learn these two tasks with a shared encoder. En-to-De and En-to-Es experiments on the MuSTC dataset demonstrate that our proposed technique achieves substantially better translation quality at similar levels of latency.

Renjie Zheng, Junkun Chen, Mingbo Ma et al.

Recently, representation learning for text and speech has successfully improved many language related tasks. However, all existing methods suffer from two limitations: (a) they only learn from one input modality, while a unified representation for both speech and text is needed by tasks such as end-to-end speech translation, and as a result,(b) they can not exploit various large-scale text and speech data and their performance is limited by the scarcity of parallel speech translation data.To address these problems, we propose a Fused Acoustic and Text Masked Language Model (FAT-MLM) which jointly learns a unified representation for both acoustic and text input from various types of corpora including parallel data for speech recognition and machine translation, and even pure speech and text data. Within this cross-modal representation learning framework, we further present an end-to-end model for Fused Acoustic and Text Speech Translation (FAT-ST). Experiments on three translation directions show that by fine-tuning from FAT-MLM, our proposed speech translation models substantially improve translation quality by up to +5.9 BLEU.

Jingbo Zhou, Shuangli Li, Liang Huang et al.

Accurately predicting the binding affinity between drugs and proteins is an essential step for computational drug discovery. Since graph neural networks (GNNs) have demonstrated remarkable success in various graph-related tasks, GNNs have been considered as a promising tool to improve the binding affinity prediction in recent years. However, most of the existing GNN architectures can only encode the topological graph structure of drugs and proteins without considering the relative spatial information among their atoms. Whereas, different from other graph datasets such as social networks and commonsense knowledge graphs, the relative spatial position and chemical bonds among atoms have significant impacts on the binding affinity. To this end, in this paper, we propose a diStance-aware Molecule graph Attention Network (S-MAN) tailored to drug-target binding affinity prediction. As a dedicated solution, we first propose a position encoding mechanism to integrate the topological structure and spatial position information into the constructed pocket-ligand graph. Moreover, we propose a novel edge-node hierarchical attentive aggregation structure which has edge-level aggregation and node-level aggregation. The hierarchical attentive aggregation can capture spatial dependencies among atoms, as well as fuse the position-enhanced information with the capability of discriminating multiple spatial relations among atoms. Finally, we conduct extensive experiments on two standard datasets to demonstrate the effectiveness of S-MAN.

Xiangci Li, Hairong Liu, Liang Huang

Existing natural language processing systems are vulnerable to noisy inputs resulting from misspellings. On the contrary, humans can easily infer the corresponding correct words from their misspellings and surrounding context. Inspired by this, we address the stand-alone spelling correction problem, which only corrects the spelling of each token without additional token insertion or deletion, by utilizing both spelling information and global context representations. We present a simple yet powerful solution that jointly detects and corrects misspellings as a sequence labeling task by fine-turning a pre-trained language model. Our solution outperforms the previous state-of-the-art result by 12.8% absolute F0.5 score.

Zhuangzhi Chen, Hui Cui, Jingyang Xiang et al.

Deep learning methods achieve great success in many areas due to their powerful feature extraction capabilities and end-to-end training mechanism, and recently they are also introduced for radio signal modulation classification. In this paper, we propose a novel deep learning framework called SigNet, where a signal-to-matrix (S2M) operator is adopted to convert the original signal into a square matrix first and is co-trained with a follow-up CNN architecture for classification. This model is further accelerated by integrating 1D convolution operators, leading to the upgraded model SigNet2.0. The simulations on two signal datasets show that both SigNet and SigNet2.0 outperform a number of well-known baselines. More interestingly, our proposed models behave extremely well in small-sample learning when only a small training dataset is provided. They can achieve a relatively high accuracy even when 1\% training data are kept, while other baseline models may lose their effectiveness much more quickly as the datasets get smaller. Such result suggests that SigNet/SigNet2.0 could be extremely useful in the situations where labeled signal data are difficult to obtain. The visualization of the output features of our models demonstrates that our model can well divide different modulation types of signals in the feature hyper-space.

Junkun Chen, Mingbo Ma, Renjie Zheng et al.

End-to-end Speech-to-text Translation (E2E-ST), which directly translates source language speech to target language text, is widely useful in practice, but traditional cascaded approaches (ASR+MT) often suffer from error propagation in the pipeline. On the other hand, existing end-to-end solutions heavily depend on the source language transcriptions for pre-training or multi-task training with Automatic Speech Recognition (ASR). We instead propose a simple technique to learn a robust speech encoder in a self-supervised fashion only on the speech side, which can utilize speech data without transcription. This technique termed Masked Acoustic Modeling (MAM), not only provides an alternative solution to improving E2E-ST, but also can perform pre-training on any acoustic signals (including non-speech ones) without annotation. We conduct our experiments over 8 different translation directions. In the setting without using any transcriptions, our technique achieves an average improvement of +1.1 BLEU, and +2.3 BLEU with MAM pre-training. Pre-training of MAM with arbitrary acoustic signals also has an average improvement with +1.6 BLEU for those languages. Compared with ASR multi-task learning solution, which replies on transcription during training, our pre-trained MAM model, which does not use transcription, achieves similar accuracy.

Junkun Chen, Renjie Zheng, Atsuhito Kita et al.

Simultaneous translation is vastly different from full-sentence translation, in the sense that it starts translation before the source sentence ends, with only a few words delay. However, due to the lack of large-scale, high-quality simultaneous translation datasets, most such systems are still trained on conventional full-sentence bitexts. This is far from ideal for the simultaneous scenario due to the abundance of unnecessary long-distance reorderings in those bitexts. We propose a novel method that rewrites the target side of existing full-sentence corpora into simultaneous-style translation. Experiments on Zh->En and Ja->En simultaneous translation show substantial improvements (up to +2.7 BLEU) with the addition of these generated pseudo-references.

Renjie Zheng, Mingbo Ma, Baigong Zheng et al.

Simultaneous speech-to-speech translation is widely useful but extremely challenging, since it needs to generate target-language speech concurrently with the source-language speech, with only a few seconds delay. In addition, it needs to continuously translate a stream of sentences, but all recent solutions merely focus on the single-sentence scenario. As a result, current approaches accumulate latencies progressively when the speaker talks faster, and introduce unnatural pauses when the speaker talks slower. To overcome these issues, we propose Self-Adaptive Translation (SAT) which flexibly adjusts the length of translations to accommodate different source speech rates. At similar levels of translation quality (as measured by BLEU), our method generates more fluent target speech (as measured by the naturalness metric MOS) with substantially lower latency than the baseline, in both Zh <-> En directions.

Chunhui Zhou, Shangding Gu, Yuanqiao Wen et al.

The essence of the path planning problems is multi-modality constraint. However, most of the current literature has not mentioned this issue. This paper introduces the research progress of path planning based on the multi-modality constraint. The path planning of multi-modality constraint research can be classified into three stages in terms of its basic ingredients (such as shape, kinematics and dynamics et al.): Route Planning, Trajectory Planning and Motion Planning. It then reviews the research methods and classical algorithms, especially those applied to the Unmanned Surface Vehicle (USV) in every stage. Finally, the paper points out some existing problems in every stage and suggestions for future research.

Boxiang Liu, Liang Huang

Machine translation requires large amounts of parallel text. While such datasets are abundant in domains such as newswire, they are less accessible in the biomedical domain. Chinese and English are two of the most widely spoken languages, yet to our knowledge a parallel corpus in the biomedical domain does not exist for this language pair. In this study, we develop an effective pipeline to acquire and process an English-Chinese parallel corpus, consisting of about 100,000 sentence pairs and 3,000,000 tokens on each side, from the New England Journal of Medicine (NEJM). We show that training on out-of-domain data and fine-tuning with as few as 4,000 NEJM sentence pairs improve translation quality by 25.3 (13.4) BLEU for en$\to$zh (zh$\to$en) directions. Translation quality continues to improve at a slower pace on larger in-domain datasets, with an increase of 33.0 (24.3) BLEU for en$\to$zh (zh$\to$en) directions on the full dataset.

Liang Huang, You Zhang, Weijian Pan et al.

Deep learning has recently been successfully applied in automatic modulation classification by extracting and classifying radio features in an end-to-end way. However, deep learning-based radio modulation classifiers are lack of interpretability, and there is little explanation or visibility into what kinds of radio features are extracted and chosen for classification. In this paper, we visualize different deep learning-based radio modulation classifiers by introducing a class activation vector. Specifically, both convolutional neural networks (CNN) based classifier and long short-term memory (LSTM) based classifier are separately studied, and their extracted radio features are visualized. Extensive numerical results show both the CNN-based classifier and LSTM-based classifier extract similar radio features relating to modulation reference points. In particular, for the LSTM-based classifier, its obtained radio features are similar to the knowledge of human experts. Our numerical results indicate the radio features extracted by deep learning-based classifiers greatly depend on the contents carried by radio signals, and a short radio sample may lead to misclassification.

Renjie Zheng, Mingbo Ma, Baigong Zheng et al.

Simultaneous translation has many important application scenarios and attracts much attention from both academia and industry recently. Most existing frameworks, however, have difficulties in balancing between the translation quality and latency, i.e., the decoding policy is usually either too aggressive or too conservative. We propose an opportunistic decoding technique with timely correction ability, which always (over-)generates a certain mount of extra words at each step to keep the audience on track with the latest information. At the same time, it also corrects, in a timely fashion, the mistakes in the former overgenerated words when observing more source context to ensure high translation quality. Experiments show our technique achieves substantial reduction in latency and up to +3.1 increase in BLEU, with revision rate under 8% in Chinese-to-English and English-to-Chinese translation.

Baigong Zheng, Kaibo Liu, Renjie Zheng et al.

Adaptive policies are better than fixed policies for simultaneous translation, since they can flexibly balance the tradeoff between translation quality and latency based on the current context information. But previous methods on obtaining adaptive policies either rely on complicated training process, or underperform simple fixed policies. We design an algorithm to achieve adaptive policies via a simple heuristic composition of a set of fixed policies. Experiments on Chinese -> English and German -> English show that our adaptive policies can outperform fixed ones by up to 4 BLEU points for the same latency, and more surprisingly, it even surpasses the BLEU score of full-sentence translation in the greedy mode (and very close to beam mode), but with much lower latency.

Liang Huang, Weijian Pan, You Zhang et al.

Deep learning has recently been applied to automatically classify the modulation categories of received radio signals without manual experience. However, training deep learning models requires massive volume of data. An insufficient training data will cause serious overfitting problem and degrade the classification accuracy. To cope with small dataset, data augmentation has been widely used in image processing to expand the dataset and improve the robustness of deep learning models. However, in wireless communication areas, the effect of different data augmentation methods on radio modulation classification has not been studied yet. In this paper, we evaluate different data augmentation methods via a state-of-the-art deep learning-based modulation classifier. Based on the characteristics of modulated signals, three augmentation methods are considered, i.e., rotation, flip, and Gaussian noise, which can be applied in both training phase and inference phase of the deep learning algorithm. Numerical results show that all three augmentation methods can improve the classification accuracy. Among which, the rotation augmentation method outperforms the flip method, both of which achieve higher classification accuracy than the Gaussian noise method. Given only 12.5% of training dataset, a joint rotation and flip augmentation policy can achieve even higher classification accuracy than the baseline with initial 100% training dataset without augmentation. Furthermore, with data augmentation, radio modulation categories can be successfully classified using shorter radio samples, leading to a simplified deep learning model and shorter the classification response time.

Mingbo Ma, Baigong Zheng, Kaibo Liu et al.

Text-to-speech synthesis (TTS) has witnessed rapid progress in recent years, where neural methods became capable of producing audios with high naturalness. However, these efforts still suffer from two types of latencies: (a) the {\em computational latency} (synthesizing time), which grows linearly with the sentence length even with parallel approaches, and (b) the {\em input latency} in scenarios where the input text is incrementally generated (such as in simultaneous translation, dialog generation, and assistive technologies). To reduce these latencies, we devise the first neural incremental TTS approach based on the recently proposed prefix-to-prefix framework. We synthesize speech in an online fashion, playing a segment of audio while generating the next, resulting in an $O(1)$ rather than $O(n)$ latency.

Hairong Liu, Mingbo Ma, Liang Huang

The research in machine translation community focus on translation in text space. However, humans are in fact also good at direct translation in pronunciation space. Some existing translation systems, such as simultaneous machine translation, are inherently more natural and thus potentially more robust by directly translating in pronunciation space. In this paper, we conduct large scale experiments on a self-built dataset with about $20$M En-Zh pairs of text sentences and corresponding pronunciation sentences. We proposed three new categories of translations: $1)$ translating a pronunciation sentence in source language into a pronunciation sentence in target language (P2P-Tran), $2)$ translating a text sentence in source language into a pronunciation sentence in target language (T2P-Tran), and $3)$ translating a pronunciation sentence in source language into a text sentence in target language (P2T-Tran), and compare them with traditional text translation (T2T-Tran). Our experiments clearly show that all $4$ categories of translations have comparable performances, with small and sometimes ignorable differences.

Renjie Zheng, Mingbo Ma, Baigong Zheng et al.

Beam search is universally used in full-sentence translation but its application to simultaneous translation remains non-trivial, where output words are committed on the fly. In particular, the recently proposed wait-k policy (Ma et al., 2019a) is a simple and effective method that (after an initial wait) commits one output word on receiving each input word, making beam search seemingly impossible. To address this challenge, we propose a speculative beam search algorithm that hallucinates several steps into the future in order to reach a more accurate decision, implicitly benefiting from a target language model. This makes beam search applicable for the first time to the generation of a single word in each step. Experiments over diverse language pairs show large improvements over previous work.

Baigong Zheng, Renjie Zheng, Mingbo Ma et al.

Simultaneous translation is widely useful but remains challenging. Previous work falls into two main categories: (a) fixed-latency policies such as Ma et al. (2019) and (b) adaptive policies such as Gu et al. (2017). The former are simple and effective, but have to aggressively predict future content due to diverging source-target word order; the latter do not anticipate, but suffer from unstable and inefficient training. To combine the merits of both approaches, we propose a simple supervised-learning framework to learn an adaptive policy from oracle READ/WRITE sequences generated from parallel text. At each step, such an oracle sequence chooses to WRITE the next target word if the available source sentence context provides enough information to do so, otherwise READ the next source word. Experiments on German<->English show that our method, without retraining the underlying NMT model, can learn flexible policies with better BLEU scores and similar latencies compared to previous work.

Renjie Zheng, Hairong Liu, Mingbo Ma et al.

This paper describes the machine translation system developed jointly by Baidu Research and Oregon State University for WMT 2019 Machine Translation Robustness Shared Task. Translation of social media is a very challenging problem, since its style is very different from normal parallel corpora (e.g. News) and also include various types of noises. To make it worse, the amount of social media parallel corpora is extremely limited. In this paper, we use a domain sensitive training method which leverages a large amount of parallel data from popular domains together with a little amount of parallel data from social media. Furthermore, we generate a parallel dataset with pseudo noisy source sentences which are back-translated from monolingual data using a model trained by a similar domain sensitive way. We achieve more than 10 BLEU improvement in both En-Fr and Fr-En translation compared with the baseline methods.

Baigong Zheng, Renjie Zheng, Mingbo Ma et al.

Simultaneous translation is widely useful but remains one of the most difficult tasks in NLP. Previous work either uses fixed-latency policies, or train a complicated two-staged model using reinforcement learning. We propose a much simpler single model that adds a `delay' token to the target vocabulary, and design a restricted dynamic oracle to greatly simplify training. Experiments on Chinese<->English simultaneous translation show that our work leads to flexible policies that achieve better BLEU scores and lower latencies compared to both fixed and RL-learned policies.

Mingbo Ma, Renjie Zheng, Liang Huang

Beam search optimization resolves many issues in neural machine translation. However, this method lacks principled stopping criteria and does not learn how to stop during training, and the model naturally prefers the longer hypotheses during the testing time in practice since they use the raw score instead of the probability-based score. We propose a novel ranking method which enables an optimal beam search stopping criteria. We further introduce a structured prediction loss function which penalizes suboptimal finished candidates produced by beam search during training. Experiments of neural machine translation on both synthetic data and real languages (German-to-English and Chinese-to-English) demonstrate our proposed methods lead to better length and BLEU score.

Mingbo Ma, Liang Huang, Hao Xiong et al.

Simultaneous translation, which translates sentences before they are finished, is useful in many scenarios but is notoriously difficult due to word-order differences. While the conventional seq-to-seq framework is only suitable for full-sentence translation, we propose a novel prefix-to-prefix framework for simultaneous translation that implicitly learns to anticipate in a single translation model. Within this framework, we present a very simple yet surprisingly effective wait-k policy trained to generate the target sentence concurrently with the source sentence, but always k words behind. Experiments show our strategy achieves low latency and reasonable quality (compared to full-sentence translation) on 4 directions: zh<->en and de<->en.

Hairong Liu, Mingbo Ma, Liang Huang et al.

Neural machine translation (NMT) is notoriously sensitive to noises, but noises are almost inevitable in practice. One special kind of noise is the homophone noise, where words are replaced by other words with similar pronunciations. We propose to improve the robustness of NMT to homophone noises by 1) jointly embedding both textual and phonetic information of source sentences, and 2) augmenting the training dataset with homophone noises. Interestingly, to achieve better translation quality and more robustness, we found that most (though not all) weights should be put on the phonetic rather than textual information. Experiments show that our method not only significantly improves the robustness of NMT to homophone noises, but also surprisingly improves the translation quality on some clean test sets.

Wen Zhang, Liang Huang, Yang Feng et al.

Although neural machine translation has achieved promising results, it suffers from slow translation speed. The direct consequence is that a trade-off has to be made between translation quality and speed, thus its performance can not come into full play. We apply cube pruning, a popular technique to speed up dynamic programming, into neural machine translation to speed up the translation. To construct the equivalence class, similar target hidden states are combined, leading to less RNN expansion operations on the target side and less \$\mathrm{softmax}\$ operations over the large target vocabulary. The experiments show that, at the same or even better translation quality, our method can translate faster compared with naive beam search by \$3.3\times\$ on GPUs and \$3.5\times\$ on CPUs.

Liang Huang, Kai Zhao, Mingbo Ma

In neural text generation such as neural machine translation, summarization, and image captioning, beam search is widely used to improve the output text quality. However, in the neural generation setting, hypotheses can finish in different steps, which makes it difficult to decide when to end beam search to ensure optimality. We propose a provably optimal beam search algorithm that will always return the optimal-score complete hypothesis (modulo beam size), and finish as soon as the optimality is established (finishing no later than the baseline). To counter neural generation's tendency for shorter hypotheses, we also introduce a bounded length reward mechanism which allows a modified version of our beam search algorithm to remain optimal. Experiments on neural machine translation demonstrate that our principled beam search algorithm leads to improvement in BLEU score over previously proposed alternatives.

Renjie Zheng, Yilin Yang, Mingbo Ma et al.

This paper describes multimodal machine translation systems developed jointly by Oregon State University and Baidu Research for WMT 2018 Shared Task on multimodal translation. In this paper, we introduce a simple approach to incorporate image information by feeding image features to the decoder side. We also explore different sequence level training methods including scheduled sampling and reinforcement learning which lead to substantial improvements. Our systems ensemble several models using different architectures and training methods and achieve the best performance for three subtasks: En-De and En-Cs in task 1 and (En+De+Fr)-Cs task 1B.

Yilin Yang, Liang Huang, Mingbo Ma

Beam search is widely used in neural machine translation, and usually improves translation quality compared to greedy search. It has been widely observed that, however, beam sizes larger than 5 hurt translation quality. We explain why this happens, and propose several methods to address this problem. Furthermore, we discuss the optimal stopping criteria for these methods. Results show that our hyperparameter-free methods outperform the widely-used hyperparameter-free heuristic of length normalization by +2.0 BLEU, and achieve the best results among all methods on Chinese-to-English translation.

Renjie Zheng, Mingbo Ma, Liang Huang

Neural text generation, including neural machine translation, image captioning, and summarization, has been quite successful recently. However, during training time, typically only one reference is considered for each example, even though there are often multiple references available, e.g., 4 references in NIST MT evaluations, and 5 references in image captioning data. We first investigate several different ways of utilizing multiple human references during training. But more importantly, we then propose an algorithm to generate exponentially many pseudo-references by first compressing existing human references into lattices and then traversing them to generate new pseudo-references. These approaches lead to substantial improvements over strong baselines in both machine translation (+1.5 BLEU) and image captioning (+3.1 BLEU / +11.7 CIDEr).

Jiaji Huang, Yi Li, Wei Ping et al.

We propose a large margin criterion for training neural language models. Conventionally, neural language models are trained by minimizing perplexity (PPL) on grammatical sentences. However, we demonstrate that PPL may not be the best metric to optimize in some tasks, and further propose a large margin formulation. The proposed method aims to enlarge the margin between the "good" and "bad" sentences in a task-specific sense. It is trained end-to-end and can be widely applied to tasks that involve re-scoring of generated text. Compared with minimum-PPL training, our method gains up to 1.1 WER reduction for speech recognition and 1.0 BLEU increase for machine translation.

Juneki Hong, Liang Huang

Recently, span-based constituency parsing has achieved competitive accuracies with extremely simple models by using bidirectional RNNs to model "spans". However, the minimal span parser of Stern et al (2017a) which holds the current state of the art accuracy is a chart parser running in cubic time, $O(n^3)$, which is too slow for longer sentences and for applications beyond sentence boundaries such as end-to-end discourse parsing and joint sentence boundary detection and parsing. We propose a linear-time constituency parser with RNNs and dynamic programming using graph-structured stack and beam search, which runs in time $O(n b^2)$ where $b$ is the beam size. We further speed this up to $O(n b\log b)$ by integrating cube pruning. Compared with chart parsing baselines, this linear-time parser is substantially faster for long sentences on the Penn Treebank and orders of magnitude faster for discourse parsing, and achieves the highest F1 accuracy on the Penn Treebank among single model end-to-end systems.

Reza Ghaeini, Xiaoli Z. Fern, Liang Huang et al.

Traditional event detection methods heavily rely on manually engineered rich features. Recent deep learning approaches alleviate this problem by automatic feature engineering. But such efforts, like tradition methods, have so far only focused on single-token event mentions, whereas in practice events can also be a phrase. We instead use forward-backward recurrent neural networks (FBRNNs) to detect events that can be either words or phrases. To the best our knowledge, this is one of the first efforts to handle multi-word events and also the first attempt to use RNNs for event detection. Experimental results demonstrate that FBRNN is competitive with the state-of-the-art methods on the ACE 2005 and the Rich ERE 2015 event detection tasks.