Yang Li

Qinkai Zheng, Xiao Xia, Xu Zou et al.

Large pre-trained code generation models, such as OpenAI Codex, can generate syntax- and function-correct code, making the coding of programmers more productive and our pursuit of artificial general intelligence closer. In this paper, we introduce CodeGeeX, a multilingual model with 13 billion parameters for code generation. CodeGeeX is pre-trained on 850 billion tokens of 23 programming languages as of June 2022. Our extensive experiments suggest that CodeGeeX outperforms multilingual code models of similar scale for both the tasks of code generation and translation on HumanEval-X. Building upon HumanEval (Python only), we develop the HumanEval-X benchmark for evaluating multilingual models by hand-writing the solutions in C++, Java, JavaScript, and Go. In addition, we build CodeGeeX-based extensions on Visual Studio Code, JetBrains, and Cloud Studio, generating 4.7 billion tokens for tens of thousands of active users per week. Our user study demonstrates that CodeGeeX can help to increase coding efficiency for 83.4% of its users. Finally, CodeGeeX is publicly accessible and in Sep. 2022, we open-sourced its code, model weights (the version of 850B tokens), API, extensions, and HumanEval-X at https://github.com/THUDM/CodeGeeX.

Xiaocui Yang, Wenfang Wu, Shi Feng et al.

The popularity of multimodal large language models (MLLMs) has triggered a recent surge in research efforts dedicated to evaluating these models. Nevertheless, existing evaluation studies of MLLMs primarily focus on the comprehension and reasoning of unimodal (vision) content, neglecting performance evaluations in the domain of multimodal (vision-language) content understanding. Beyond multimodal reasoning, tasks related to multimodal content comprehension necessitate a profound understanding of multimodal contexts, achieved through the multimodal interaction to obtain a final answer. In this paper, we introduce a comprehensive assessment framework called MM-BigBench, which incorporates a diverse range of metrics to offer an extensive evaluation of the performance of various models and instructions across a wide spectrum of diverse multimodal content comprehension tasks. Consequently, our work complements research on the performance of MLLMs in multimodal comprehension tasks, achieving a more comprehensive and holistic evaluation of MLLMs. To begin, we employ the Best Performance metric to ascertain each model's performance upper bound on different datasets. Subsequently, the Mean Relative Gain metric offers an assessment of the overall performance of various models and instructions, while the Stability metric measures their sensitivity. Furthermore, previous research centers on evaluating models independently or solely assessing instructions, neglecting the adaptability between models and instructions. We propose the Adaptability metric to quantify the adaptability between models and instructions. Our paper evaluates a total of 20 language models (14 MLLMs) on 14 multimodal datasets spanning 6 tasks, with 10 instructions for each task, and derives novel insights. Our code will be released at https://github.com/declare-lab/MM-BigBench.

Junyi Lu, Lei Yu, Xiaojia Li et al.

The automation of code review activities, a long-standing pursuit in software engineering, has been primarily addressed by numerous domain-specific pre-trained models. Despite their success, these models frequently demand extensive resources for pre-training from scratch. In contrast, Large Language Models (LLMs) provide an intriguing alternative, given their remarkable capabilities when supplemented with domain-specific knowledge. However, their potential for automating code review tasks remains largely unexplored. In response to this research gap, we present LLaMA-Reviewer, an innovative framework that leverages the capabilities of LLaMA, a popular LLM, in the realm of code review. Mindful of resource constraints, this framework employs parameter-efficient fine-tuning (PEFT) methods, delivering high performance while using less than 1% of trainable parameters. An extensive evaluation of LLaMA-Reviewer is conducted on two diverse, publicly available datasets. Notably, even with the smallest LLaMA base model consisting of 6.7B parameters and a limited number of tuning epochs, LLaMA-Reviewer equals the performance of existing code-review-focused models. The ablation experiments provide insights into the influence of various fine-tuning process components, including input representation, instruction tuning, and different PEFT methods. To foster continuous progress in this field, the code and all PEFT-weight plugins have been made open-source.

Bowen Jin, Hansi Zeng, Guoyin Wang et al.

Semantic identifier (ID) is an important concept in information retrieval that aims to preserve the semantics of objects such as documents and items inside their IDs. Previous studies typically adopt a two-stage pipeline to learn semantic IDs by first procuring embeddings using off-the-shelf text encoders and then deriving IDs based on the embeddings. However, each step introduces potential information loss, and there is usually an inherent mismatch between the distribution of embeddings within the latent space produced by text encoders and the anticipated distribution required for semantic indexing. It is non-trivial to design a method that can learn the document's semantic representations and its hierarchical structure simultaneously, given that semantic IDs are discrete and sequentially structured, and the semantic supervision is deficient. In this paper, we introduce LMIndexer, a self-supervised framework to learn semantic IDs with a generative language model. We tackle the challenge of sequential discrete ID by introducing a semantic indexer capable of generating neural sequential discrete representations with progressive training and contrastive learning. In response to the semantic supervision deficiency, we propose to train the model with a self-supervised document reconstruction objective. We show the high quality of the learned IDs and demonstrate their effectiveness on three tasks including recommendation, product search, and document retrieval on five datasets from various domains. Code is available at https://github.com/PeterGriffinJin/LMIndexer.

Guanzi Chen, Jiying Zhang, Xi Xiao et al.

Recently, test time adaptation (TTA) has attracted increasing attention due to its power of handling the distribution shift issue in the real world. Unlike what has been developed for convolutional neural networks (CNNs) for image data, TTA is less explored for Graph Neural Networks (GNNs). There is still a lack of efficient algorithms tailored for graphs with irregular structures. In this paper, we present a novel test time adaptation strategy named Graph Adversarial Pseudo Group Contrast (GAPGC), for graph neural networks TTA, to better adapt to the Out Of Distribution (OOD) test data. Specifically, GAPGC employs a contrastive learning variant as a self-supervised task during TTA, equipped with Adversarial Learnable Augmenter and Group Pseudo-Positive Samples to enhance the relevance between the self-supervised task and the main task, boosting the performance of the main task. Furthermore, we provide theoretical evidence that GAPGC can extract minimal sufficient information for the main task from information theory perspective. Extensive experiments on molecular scaffold OOD dataset demonstrated that the proposed approach achieves state-of-the-art performance on GNNs.

Yang Li, Canran Xu, Guodong Long et al.

Recently, prefix-tuning was proposed to efficiently adapt pre-trained language models to a broad spectrum of natural language classification tasks. It leverages soft prefix as task-specific indicators and language verbalizers as categorical-label mentions to narrow the formulation gap from pre-training language models. However, when the label space increases considerably (i.e., many-class classification), such a tuning technique suffers from a verbalizer ambiguity problem since the many-class labels are represented by semantic-similar verbalizers in short language phrases. To overcome this, inspired by the human-decision process that the most ambiguous classes would be mulled over for each instance, we propose a brand-new prefix-tuning method, Counterfactual Contrastive Prefix-tuning (CCPrefix), for many-class classification. Basically, an instance-dependent soft prefix, derived from fact-counterfactual pairs in the label space, is leveraged to complement the language verbalizers in many-class classification. We conduct experiments on many-class benchmark datasets in both the fully supervised setting and the few-shot setting, which indicates that our model outperforms former baselines.

Yang Li, Shao Zhang, Jichen Sun et al.

Zero-shot coordination in cooperative artificial intelligence (AI) remains a significant challenge, which means effectively coordinating with a wide range of unseen partners. Previous algorithms have attempted to address this challenge by optimizing fixed objectives within a population to improve strategy or behaviour diversity. However, these approaches can result in a loss of learning and an inability to cooperate with certain strategies within the population, known as cooperative incompatibility. To address this issue, we propose the Cooperative Open-ended LEarning (COLE) framework, which constructs open-ended objectives in cooperative games with two players from the perspective of graph theory to assess and identify the cooperative ability of each strategy. We further specify the framework and propose a practical algorithm that leverages knowledge from game theory and graph theory. Furthermore, an analysis of the learning process of the algorithm shows that it can efficiently overcome cooperative incompatibility. The experimental results in the Overcooked game environment demonstrate that our method outperforms current state-of-the-art methods when coordinating with different-level partners. Our demo is available at https://sites.google.com/view/cole-2023.

Jiyong Li, Dilshod Azizov, Yang Li et al.

Recently, because of the high-quality representations of contrastive learning methods, rehearsal-based contrastive continual learning has been proposed to explore how to continually learn transferable representation embeddings to avoid the catastrophic forgetting issue in traditional continual settings. Based on this framework, we propose Contrastive Continual Learning via Importance Sampling (CCLIS) to preserve knowledge by recovering previous data distributions with a new strategy for Replay Buffer Selection (RBS), which minimize estimated variance to save hard negative samples for representation learning with high quality. Furthermore, we present the Prototype-instance Relation Distillation (PRD) loss, a technique designed to maintain the relationship between prototypes and sample representations using a self-distillation process. Experiments on standard continual learning benchmarks reveal that our method notably outperforms existing baselines in terms of knowledge preservation and thereby effectively counteracts catastrophic forgetting in online contexts. The code is available at https://github.com/lijy373/CCLIS.

Junyi Lu, Xiaojia Li, Zihan Hua et al.

Code review is a vital but demanding aspect of software development, generating significant interest in automating review comments. Traditional evaluation methods for these comments, primarily based on text similarity, face two major challenges: inconsistent reliability of human-authored comments in open-source projects and the weak correlation of text similarity with objectives like enhancing code quality and detecting defects. This study empirically analyzes benchmark comments using a novel set of criteria informed by prior research and developer interviews. We then similarly revisit the evaluation of existing methodologies. Our evaluation framework, DeepCRCEval, integrates human evaluators and Large Language Models (LLMs) for a comprehensive reassessment of current techniques based on the criteria set. Besides, we also introduce an innovative and efficient baseline, LLM-Reviewer, leveraging the few-shot learning capabilities of LLMs for a target-oriented comparison. Our research highlights the limitations of text similarity metrics, finding that less than 10% of benchmark comments are high quality for automation. In contrast, DeepCRCEval effectively distinguishes between high and low-quality comments, proving to be a more reliable evaluation mechanism. Incorporating LLM evaluators into DeepCRCEval significantly boosts efficiency, reducing time and cost by 88.78% and 90.32%, respectively. Furthermore, LLM-Reviewer demonstrates significant potential of focusing task real targets in comment generation.

Yang Li, Kangbo Liu, Yaoxin Wu et al.

Bundle recommendations strive to offer users a set of items as a package named bundle, enhancing convenience and contributing to the seller's revenue. While previous approaches have demonstrated notable performance, we argue that they may compromise the ternary relationship among users, items, and bundles. This compromise can result in information loss, ultimately impacting the overall model performance. To address this gap, we develop a unified model for bundle recommendation, termed hypergraph-enhanced dual convolutional neural network (HED). Our approach is characterized by two key aspects. Firstly, we construct a complete hypergraph to capture interaction dynamics among users, items, and bundles. Secondly, we incorporate U-B interaction information to enhance the information representation derived from users and bundle embedding vectors. Extensive experimental results on the Youshu and Netease datasets have demonstrated that HED surpasses state-of-the-art baselines, proving its effectiveness. In addition, various ablation studies and sensitivity analyses revealed the working mechanism and proved our effectiveness. Codes and datasets are available at https://github.com/AAI-Lab/HED

Lu Ye, Ze Tao, Yong Huang et al.

Self-attention is an essential component of large language models (LLM) but a significant source of inference latency for long sequences. In multi-tenant LLM serving scenarios, the compute and memory operation cost of self-attention can be optimized by using the probability that multiple LLM requests have shared system prompts in prefixes. In this paper, we introduce ChunkAttention, a prefix-aware self-attention module that can detect matching prompt prefixes across multiple requests and share their key/value tensors in memory at runtime to improve the memory utilization of KV cache. This is achieved by breaking monolithic key/value tensors into smaller chunks and structuring them into the auxiliary prefix tree. Consequently, on top of the prefix-tree based KV cache, we design an efficient self-attention kernel, where a two-phase partition algorithm is implemented to improve the data locality during self-attention computation in the presence of shared system prompts. Experiments show that ChunkAttention can speed up the self-attention kernel by 3.2-4.8$\times$ compared to the state-of-the-art implementation, with the length of the system prompt ranging from 1024 to 4096.

Mingda Chen, Yang Li, Xilun Chen et al.

Long-form factuality evaluation assesses the ability of models to generate accurate, comprehensive responses to short prompts. Existing benchmarks often lack human verification, leading to potential quality issues. To address this limitation, we introduce FACTORY, a large-scale, human-verified prompt set. Developed using a model-in-the-loop approach and refined by humans, FACTORY includes challenging prompts that are fact-seeking, answerable, and unambiguous. We conduct human evaluations on 6 state-of-the-art language models using FACTORY and existing datasets. Our results show that FACTORY is a challenging benchmark: approximately 40% of the claims made in the responses of SOTA models are not factual, compared to only 10% for other datasets. Our analysis identifies the strengths of FACTORY over prior benchmarks, emphasizing its reliability and the necessity for models to reason across long-tailed facts.

Yang Li, Songlin Yang, Wei Wang et al.

Advanced diffusion-based Text-to-Image (T2I) models, such as the Stable Diffusion Model, have made significant progress in generating diverse and high-quality images using text prompts alone. However, when non-famous users require personalized image generation for their identities (IDs), the T2I models fail to accurately generate their ID-related images. The main problem is that pre-trained T2I models do not learn the mapping between the new ID prompts and their corresponding visual content. The previous methods either failed to accurately fit the face region or lost the interactive generative ability with other existing concepts in T2I models. In other words, they are unable to generate T2I-aligned and semantic-fidelity images for the given prompts with other concepts such as scenes (``Eiffel Tower''), actions (``holding a basketball''), and facial attributes (``eyes closed''). In this paper, we focus on inserting accurate and interactive ID embedding into the Stable Diffusion Model for semantic-fidelity personalized generation. We address this challenge from two perspectives: face-wise region fitting and semantic-fidelity token optimization. Specifically, we first visualize the attention overfit problem and propose a face-wise attention loss to fit the face region instead of entangling ID-unrelated information, such as face layout and background. This key trick significantly enhances the ID accuracy and interactive generative ability with other existing concepts. Then, we optimize one ID representation as multiple per-stage tokens where each token contains two disentangled features. This expansion of the textual conditioning space improves semantic-fidelity control. Extensive experiments validate that our results exhibit superior ID accuracy, text-based manipulation ability, and generalization compared to previous methods.

Chengxi Yan, Jie Jian, Yang Li

Considering the lack of a unified framework for image description and deep cultural analysis at the subject level in the field of Ancient Chinese Paintings (ACP), this study utilized the Beijing Palace Museum's ACP collections to develop a semantic model integrating the iconological theory with a new workflow for term extraction and mapping. Our findings underscore the model's effectiveness. SDM can be used to support further art-related knowledge organization and cultural exploration of ACPs.

Bingkun Chen, Shaobing Dai, Shenghua Zheng et al.

Unsupervised dialogue structure learning is an important and meaningful task in natural language processing. The extracted dialogue structure and process can help analyze human dialogue, and play a vital role in the design and evaluation of dialogue systems. The traditional dialogue system requires experts to manually design the dialogue structure, which is very costly. But through unsupervised dialogue structure learning, dialogue structure can be automatically obtained, reducing the cost of developers constructing dialogue process. The learned dialogue structure can be used to promote the dialogue generation of the downstream task system, and improve the logic and consistency of the dialogue robot's reply.In this paper, we propose a Bert-based unsupervised dialogue structure learning algorithm DSBERT (Dialogue Structure BERT). Different from the previous SOTA models VRNN and SVRNN, we combine BERT and AutoEncoder, which can effectively combine context information. In order to better prevent the model from falling into the local optimal solution and make the dialogue state distribution more uniform and reasonable, we also propose three balanced loss functions that can be used for dialogue structure learning. Experimental results show that DSBERT can generate a dialogue structure closer to the real structure, can distinguish sentences with different semantics and map them to different hidden states.

Yang Li, Guodong Long, Tao Shen et al.

Distant supervision uses triple facts in knowledge graphs to label a corpus for relation extraction, leading to wrong labeling and long-tail problems. Some works use the hierarchy of relations for knowledge transfer to long-tail relations. However, a coarse-grained relation often implies only an attribute (e.g., domain or topic) of the distant fact, making it hard to discriminate relations based solely on sentence semantics. One solution is resorting to entity types, but open questions remain about how to fully leverage the information of entity types and how to align multi-granular entity types with sentences. In this work, we propose a novel model to enrich distantly-supervised sentences with entity types. It consists of (1) a pairwise type-enriched sentence encoding module injecting both context-free and -related backgrounds to alleviate sentence-level wrong labeling, and (2) a hierarchical type-sentence alignment module enriching a sentence with the triple fact's basic attributes to support long-tail relations. Our model achieves new state-of-the-art results in overall and long-tail performance on benchmarks.

Yang Li, Tong Chen, Peng-Fei Zhang et al.

Modern deep neural networks (DNNs) have greatly facilitated the development of sequential recommender systems by achieving state-of-the-art recommendation performance on various sequential recommendation tasks. Given a sequence of interacted items, existing DNN-based sequential recommenders commonly embed each item into a unique vector to support subsequent computations of the user interest. However, due to the potentially large number of items, the over-parameterised item embedding matrix of a sequential recommender has become a memory bottleneck for efficient deployment in resource-constrained environments, e.g., smartphones and other edge devices. Furthermore, we observe that the widely-used multi-head self-attention, though being effective in modelling sequential dependencies among items, heavily relies on redundant attention units to fully capture both global and local item-item transition patterns within a sequence. In this paper, we introduce a novel lightweight self-attentive network (LSAN) for sequential recommendation. To aggressively compress the original embedding matrix, LSAN leverages the notion of compositional embeddings, where each item embedding is composed by merging a group of selected base embedding vectors derived from substantially smaller embedding matrices. Meanwhile, to account for the intrinsic dynamics of each item, we further propose a temporal context-aware embedding composition scheme. Besides, we develop an innovative twin-attention network that alleviates the redundancy of the traditional multi-head self-attention while retaining full capacity for capturing long- and short-term (i.e., global and local) item dependencies. Comprehensive experiments demonstrate that LSAN significantly advances the accuracy and memory efficiency of existing sequential recommenders.

Yang Li, Wei Zhao, Erik Cambria et al.

Routing methods in capsule networks often learn a hierarchical relationship for capsules in successive layers, but the intra-relation between capsules in the same layer is less studied, while this intra-relation is a key factor for the semantic understanding in text data. Therefore, in this paper, we introduce a new capsule network with graph routing to learn both relationships, where capsules in each layer are treated as the nodes of a graph. We investigate strategies to yield adjacency and degree matrix with three different distances from a layer of capsules, and propose the graph routing mechanism between those capsules. We validate our approach on five text classification datasets, and our findings suggest that the approach combining bottom-up routing and top-down attention performs the best. Such an approach demonstrates generalization capability across datasets. Compared to the state-of-the-art routing methods, the improvements in accuracy in the five datasets we used were 0.82, 0.39, 0.07, 1.01, and 0.02, respectively.

Guanglin Niu, Yang Li, Chengguang Tang et al.

Aiming at expanding few-shot relations' coverage in knowledge graphs (KGs), few-shot knowledge graph completion (FKGC) has recently gained more research interests. Some existing models employ a few-shot relation's multi-hop neighbor information to enhance its semantic representation. However, noise neighbor information might be amplified when the neighborhood is excessively sparse and no neighbor is available to represent the few-shot relation. Moreover, modeling and inferring complex relations of one-to-many (1-N), many-to-one (N-1), and many-to-many (N-N) by previous knowledge graph completion approaches requires high model complexity and a large amount of training instances. Thus, inferring complex relations in the few-shot scenario is difficult for FKGC models due to limited training instances. In this paper, we propose a few-shot relational learning with global-local framework to address the above issues. At the global stage, a novel gated and attentive neighbor aggregator is built for accurately integrating the semantics of a few-shot relation's neighborhood, which helps filtering the noise neighbors even if a KG contains extremely sparse neighborhoods. For the local stage, a meta-learning based TransH (MTransH) method is designed to model complex relations and train our model in a few-shot learning fashion. Extensive experiments show that our model outperforms the state-of-the-art FKGC approaches on the frequently-used benchmark datasets NELL-One and Wiki-One. Compared with the strong baseline model MetaR, our model achieves 5-shot FKGC performance improvements of 8.0% on NELL-One and 2.8% on Wiki-One by the metric Hits@10.

Yang Li, Yu Shen, Jiawei Jiang et al.

Hyperparameter optimization (HPO) is a fundamental problem in automatic machine learning (AutoML). However, due to the expensive evaluation cost of models (e.g., training deep learning models or training models on large datasets), vanilla Bayesian optimization (BO) is typically computationally infeasible. To alleviate this issue, Hyperband (HB) utilizes the early stopping mechanism to speed up configuration evaluations by terminating those badly-performing configurations in advance. This leads to two kinds of quality measurements: (1) many low-fidelity measurements for configurations that get early-stopped, and (2) few high-fidelity measurements for configurations that are evaluated without being early stopped. The state-of-the-art HB-style method, BOHB, aims to combine the benefits of both BO and HB. Instead of sampling configurations randomly in HB, BOHB samples configurations based on a BO surrogate model, which is constructed with the high-fidelity measurements only. However, the scarcity of high-fidelity measurements greatly hampers the efficiency of BO to guide the configuration search. In this paper, we present MFES-HB, an efficient Hyperband method that is capable of utilizing both the high-fidelity and low-fidelity measurements to accelerate the convergence of HPO tasks. Designing MFES-HB is not trivial as the low-fidelity measurements can be biased yet informative to guide the configuration search. Thus we propose to build a Multi- Fidelity Ensemble Surrogate (MFES) based on the generalized Product of Experts framework, which can integrate useful information from multi-fidelity measurements effectively. The empirical studies on the real-world AutoML tasks demonstrate that MFES-HB can achieve 3.3-8.9x speedups over the state-of-the-art approach - BOHB.

Bin Fu, Yunqi Qiu, Chengguang Tang et al.

Question Answering (QA) over Knowledge Base (KB) aims to automatically answer natural language questions via well-structured relation information between entities stored in knowledge bases. In order to make KBQA more applicable in actual scenarios, researchers have shifted their attention from simple questions to complex questions, which require more KB triples and constraint inference. In this paper, we introduce the recent advances in complex QA. Besides traditional methods relying on templates and rules, the research is categorized into a taxonomy that contains two main branches, namely Information Retrieval-based and Neural Semantic Parsing-based. After describing the methods of these branches, we analyze directions for future research and introduce the models proposed by the Alime team.

Rahul Jha, Keping Bi, Yang Li et al.

We describe Artemis (Annotation methodology for Rich, Tractable, Extractive, Multi-domain, Indicative Summarization), a novel hierarchical annotation process that produces indicative summaries for documents from multiple domains. Current summarization evaluation datasets are single-domain and focused on a few domains for which naturally occurring summaries can be easily found, such as news and scientific articles. These are not sufficient for training and evaluation of summarization models for use in document management and information retrieval systems, which need to deal with documents from multiple domains. Compared to other annotation methods such as Relative Utility and Pyramid, Artemis is more tractable because judges don't need to look at all the sentences in a document when making an importance judgment for one of the sentences, while providing similarly rich sentence importance annotations. We describe the annotation process in detail and compare it with other similar evaluation systems. We also present analysis and experimental results over a sample set of 532 annotated documents.

Yang Li, Jiawei Jiang, Yingxia Shao et al.

The performance of deep neural networks crucially depends on good hyperparameter configurations. Bayesian optimization is a powerful framework for optimizing the hyperparameters of DNNs. These methods need sufficient evaluation data to approximate and minimize the validation error function of hyperparameters. However, the expensive evaluation cost of DNNs leads to very few evaluation data within a limited time, which greatly reduces the efficiency of Bayesian optimization. Besides, the previous researches focus on using the complete evaluation data to conduct Bayesian optimization, and ignore the intermediate evaluation data generated by early stopping methods. To alleviate the insufficient evaluation data problem, we propose a fast hyperparameter optimization method, HOIST, that utilizes both the complete and intermediate evaluation data to accelerate the hyperparameter optimization of DNNs. Specifically, we train multiple basic surrogates to gather information from the mixed evaluation data, and then combine all basic surrogates using weighted bagging to provide an accurate ensemble surrogate. Our empirical studies show that HOIST outperforms the state-of-the-art approaches on a wide range of DNNs, including feed forward neural networks, convolutional neural networks, recurrent neural networks, and variational autoencoder.

Patrick H. Chen, Si Si, Yang Li et al.

Model compression is essential for serving large deep neural nets on devices with limited resources or applications that require real-time responses. As a case study, a state-of-the-art neural language model usually consists of one or more recurrent layers sandwiched between an embedding layer used for representing input tokens and a softmax layer for generating output tokens. For problems with a very large vocabulary size, the embedding and the softmax matrices can account for more than half of the model size. For instance, the bigLSTM model achieves state-of- the-art performance on the One-Billion-Word (OBW) dataset with around 800k vocabulary, and its word embedding and softmax matrices use more than 6GBytes space, and are responsible for over 90% of the model parameters. In this paper, we propose GroupReduce, a novel compression method for neural language models, based on vocabulary-partition (block) based low-rank matrix approximation and the inherent frequency distribution of tokens (the power-law distribution of words). The experimental results show our method can significantly outperform traditional compression methods such as low-rank approximation and pruning. On the OBW dataset, our method achieved 6.6 times compression rate for the embedding and softmax matrices, and when combined with quantization, our method can achieve 26 times compression rate, which translates to a factor of 12.8 times compression for the entire model with very little degradation in perplexity.

Yang Li, Nan Du, Samy Bengio

Existing sequence prediction methods are mostly concerned with time-independent sequences, in which the actual time span between events is irrelevant and the distance between events is simply the difference between their order positions in the sequence. While this time-independent view of sequences is applicable for data such as natural languages, e.g., dealing with words in a sentence, it is inappropriate and inefficient for many real world events that are observed and collected at unequally spaced points of time as they naturally arise, e.g., when a person goes to a grocery store or makes a phone call. The time span between events can carry important information about the sequence dependence of human behaviors. In this work, we propose a set of methods for using time in sequence prediction. Because neural sequence models such as RNN are more amenable for handling token-like input, we propose two methods for time-dependent event representation, based on the intuition on how time is tokenized in everyday life and previous work on embedding contextualization. We also introduce two methods for using next event duration as regularization for training a sequence prediction model. We discuss these methods based on recurrent neural nets. We evaluate these methods as well as baseline models on five datasets that resemble a variety of sequence prediction tasks. The experiments revealed that the proposed methods offer accuracy gain over baseline models in a range of settings.