Haoli Bai

Kai Chen, Yunhao Gou, Runhui Huang et al.

GPT-4o, an omni-modal model that enables vocal conversations with diverse emotions and tones, marks a milestone for omni-modal foundation models. However, empowering Large Language Models to perceive and generate images, texts, and speeches end-to-end with publicly available data remains challenging for the open-source community. Existing vision-language models rely on external tools for speech processing, while speech-language models still suffer from limited or totally without vision-understanding capabilities. To address this gap, we propose the EMOVA (EMotionally Omni-present Voice Assistant), to enable Large Language Models with end-to-end speech abilities while maintaining the leading vision-language performance. With a semantic-acoustic disentangled speech tokenizer, we surprisingly notice that omni-modal alignment can further enhance vision-language and speech abilities compared with the bi-modal aligned counterparts. Moreover, a lightweight style module is introduced for the flexible speech style controls including emotions and pitches. For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks, and meanwhile, supporting omni-modal spoken dialogue with vivid emotions.

Haoli Bai, Zhiguang Liu, Xiaojun Meng et al.

Unsupervised pre-training on millions of digital-born or scanned documents has shown promising advances in visual document understanding~(VDU). While various vision-language pre-training objectives are studied in existing solutions, the document textline, as an intrinsic granularity in VDU, has seldom been explored so far. A document textline usually contains words that are spatially and semantically correlated, which can be easily obtained from OCR engines. In this paper, we propose Wukong-Reader, trained with new pre-training objectives to leverage the structural knowledge nested in document textlines. We introduce textline-region contrastive learning to achieve fine-grained alignment between the visual regions and texts of document textlines. Furthermore, masked region modeling and textline-grid matching are also designed to enhance the visual and layout representations of textlines. Experiments show that our Wukong-Reader has superior performance on various VDU tasks such as information extraction. The fine-grained alignment over textlines also empowers Wukong-Reader with promising localization ability.

Kaican Li, Lewei Yao, Jiannan Wu et al.

The ability for AI agents to "think with images" requires a sophisticated blend of reasoning and perception. However, current open multimodal agents still largely fall short on the reasoning aspect crucial for real-world tasks like analyzing documents with dense charts/diagrams and navigating maps. To address this gap, we introduce O3-Bench, a new benchmark designed to evaluate multimodal reasoning with interleaved attention to visual details. O3-Bench features challenging problems that require agents to piece together subtle visual information from distinct image areas through multi-step reasoning. The problems are highly challenging even for frontier systems like OpenAI o3, which only obtains 40.8% accuracy on O3-Bench. To make progress, we propose InSight-o3, a multi-agent framework consisting of a visual reasoning agent (vReasoner) and a visual search agent (vSearcher) for which we introduce the task of generalized visual search -- locating relational, fuzzy, or conceptual regions described in free-form language, beyond just simple objects or figures in natural images. We then present a multimodal LLM purpose-trained for this task via reinforcement learning. As a plug-and-play agent, our vSearcher empowers frontier multimodal models (as vReasoners), significantly improving their performance on a wide range of benchmarks. This marks a concrete step towards powerful o3-like open systems. Our code and dataset can be found at https://github.com/m-Just/InSight-o3 .

Ruikang Liu, Haoli Bai, Haokun Lin et al.

Large language models (LLMs) excel in natural language processing but demand intensive computation. To mitigate this, various quantization methods have been explored, yet they compromise LLM performance. This paper unveils a previously overlooked type of outliers in LLMs. Such outliers are found to allocate most of the attention scores on initial tokens of input, termed as pivot tokens, which are crucial to the performance of quantized LLMs. Given that, we propose IntactKV to generate the KV cache of pivot tokens losslessly from the full-precision model. The approach is simple and easy to combine with existing quantization solutions with no extra inference overhead. Besides, IntactKV can be calibrated as additional LLM parameters to boost the quantized LLMs further with minimal training costs. Mathematical analysis also proves that IntactKV effectively reduces the upper bound of quantization error. Empirical results show that IntactKV brings consistent improvement over various quantization methods across different LLMs and downstream tasks, leading to the new state-of-the-art for LLM quantization. The codes are available at https://github.com/ruikangliu/IntactKV.

Yuxuan Sun, Ruikang Liu, Haoli Bai et al.

Recently, quantization has been widely used for the compression and acceleration of large language models (LLMs). Due to the outliers in LLMs, it is crucial to flatten weights and activations to minimize quantization error with equally spaced quantization points. Prior research explores various pre-quantization transformations to suppress outliers, such as per-channel scaling and Hadamard transformation. However, we observe that these transformed weights and activations can still exhibit steep and dispersed distributions. In this paper, we propose FlatQuant (Fast and Learnable Affine Transformation), a new post-training quantization approach that enhances the flatness of weights and activations. Our approach identifies optimal affine transformations for each linear layer, calibrated in hours via a lightweight objective. To reduce runtime overhead of affine transformation, we apply Kronecker product with two lightweight matrices, and fuse all operations in FlatQuant into a single kernel. Extensive experiments demonstrate that FlatQuant establishes a new state-of-the-art benchmark for quantization. For example, it achieves less than 1\% accuracy drop for W4A4 quantization on the LLaMA-3-70B model, surpassing SpinQuant by 7.5\%. Additionally, it provides up to 2.3x prefill speedup and 1.7x decoding speedup compared to the FP16 model. Code is available at: https://github.com/ruikangliu/FlatQuant.

Ruikang Liu, Yuxuan Sun, Manyi Zhang et al.

Recent advancements in reasoning language models have demonstrated remarkable performance in complex tasks, but their extended chain-of-thought reasoning process increases inference overhead. While quantization has been widely adopted to reduce the inference cost of large language models, its impact on reasoning models remains understudied. In this paper, we conduct the first systematic study on quantized reasoning models, evaluating the open-sourced DeepSeek-R1-Distilled Qwen and LLaMA families ranging from 1.5B to 70B parameters, QwQ-32B, and Qwen3-8B. Our investigation covers weight, KV cache, and activation quantization using state-of-the-art algorithms at varying bit-widths, with extensive evaluation across mathematical (AIME, MATH-500), scientific (GPQA), and programming (LiveCodeBench) reasoning benchmarks. Our findings reveal that while lossless quantization can be achieved with W8A8 or W4A16 quantization, lower bit-widths introduce significant accuracy risks. We further identify model size, model origin, and task difficulty as critical determinants of performance. Contrary to expectations, quantized models do not exhibit increased output lengths. In addition, strategically scaling the model sizes or reasoning steps can effectively enhance the performance. All quantized models and codes are open-sourced in https://github.com/ruikangliu/Quantized-Reasoning-Models.

Xinrui Chen, Haoli Bai, Tao Yuan et al.

Layer pruning has emerged as a widely used technique for compressing large language models (LLMs). However, existing layer pruning approaches often incur substantial performance degradation. We identify the majority of this degradation to a single yet previously overlooked issue: \textit{the mismatch of activation magnitudes at the pruning interface}. The pre-interface activations exhibit significantly different scales from the post-interface ones, causing the distributional shift as it propagates through the remaining layers. To address this issue, we introduce \textsc{LinearPatch}, a lightweight and plug-and-play technique that fuses two operations into one matrix multiply at the pruning interface: (i) a Hadamard transformation that suppresses massive outliers at particular tokens and (ii) a channel-wise scaling that aligns activation statistics. On LLaMA-3-8B, \textsc{LinearPatch} preserves up to \textbf{94.15\%} of the original model's performance when pruning 5 out of 32 layers, outperforming the previous state of the art by \textbf{4\%}. The patch can be further refined with 5K unlabeled samples via memory-efficient offline distillation, pushing the retention to 95.16\% within only 30 minutes on a single GPU. Code is available at https://github.com/chenxinrui-tsinghua/LinearPatch.

Jing Xiong, Qiujiang Chen, Fanghua Ye et al.

Large language models (LLMs) benefit from test-time scaling but are often hampered by high inference latency. Speculative decoding is a natural way to accelerate the scaling process; however, scaling along both the parallel and sequential dimensions poses significant challenges, including substantial memory-bound execution and synchronization overhead. We introduce ATTS (Asynchronous Test-Time Scaling), a statistically guaranteed adaptive scaling framework that follows the hypothesis testing process to address these challenges. By revisiting arithmetic intensity, ATTS identifies synchronization as the primary bottleneck. It enables asynchronous inference through online calibration and proposes an ordinal classification algorithm that supports a three-stage rejection sampling pipeline, scaling along both the sequential and parallel axes. Across experiments on the MATH, AMC23, AIME24, and AIME25 datasets and across multiple draft-target model families, we show that ATTS delivers up to 56.7x speedup in test-time scaling and a 4.14x throughput improvement, while maintaining accurate control of the rejection rate, reducing latency and memory overhead, and incurring no accuracy loss. By scaling both in parallel and sequential dimensions, we enable the 1.5B/70B draft/target model combination to achieve the performance of the state-of-the-art reasoning model o3-mini (high) on the AIME dataset. We have released the code at https://github.com/menik1126/asynchronous-test-time-scaling.

Wenqian Cui, Lei Zhu, Xiaohui Li et al.

Full-Duplex Speech Language Models (FD-SLMs) are specialized foundation models designed to enable natural, real-time spoken interactions by modeling complex conversational dynamics such as interruptions, backchannels, and overlapping speech, and End-to-end (e2e) FD-SLMs leverage real-world double-channel conversational data to capture nuanced two-speaker dialogue patterns for human-like interactions. However, they face a critical challenge -- their conversational abilities often degrade compared to pure-text conversation due to prolonged speech sequences and limited high-quality spoken dialogue data. While text-guided speech generation could mitigate these issues, it suffers from timing and length issues when integrating textual guidance into double-channel audio streams, disrupting the precise time alignment essential for natural interactions. To address these challenges, we propose TurnGuide, a novel planning-inspired approach that mimics human conversational planning by dynamically segmenting assistant speech into dialogue turns and generating turn-level text guidance before speech output, which effectively resolves both insertion timing and length challenges. Extensive experiments demonstrate our approach significantly improves e2e FD-SLMs' conversational abilities, enabling them to generate semantically meaningful and coherent speech while maintaining natural conversational flow. Demos are available at https://dreamtheater123.github.io/TurnGuide-Demo/. Code will be available at https://github.com/dreamtheater123/TurnGuide.

Zhihan Guo, Feiyang Xu, Yifan Li et al.

The explosive growth in academic literature necessitates automated deep research (DR) agents, yet their evaluation remains a significant challenge. First, existing benchmarks often focus narrowly on retrieval while neglecting high-level planning and reasoning. Second, existing benchmarks favor general domains over the scientific domains that are the core application for DR agents. To address these gaps, we introduce Dr.Mi-Bench, a Modular-integrated benchmark for scientific DR agents. Grounded in academic literature, our benchmark uses a human-annotated dataset of 200 instances across 10 scientific domains, including both research and review papers. Besides, we also propose a Modular-integrated Evaluation Paradigm for DR Agents (Dr.Mi-Eval), a novel modular-integrated evaluation paradigm, which leverages the rich structure of academic papers to assess the core competencies of planning, retrieval, and reasoning through two complementary modes: an end-to-end evaluation for DR agents and an isolated evaluation for foundational LLMs as potential backbones. Experimental results reveal a fragmented performance landscape: agents exhibit specialized strengths but share critical weaknesses, most notably in performing the multi-source retrieval required for review-style tasks and performing consistently across diverse scientific fields. Moreover, improving high-level planning capability is the crucial factor for unlocking the reasoning potential of foundational LLMs as backbones. By exposing these actionable failure modes, Dr.Mi-Bench provides a diagnostic tool to guide the development of more reliable academic research assistants.

Haokun Lin, Haoli Bai, Zhili Liu et al.

Vision-language pre-trained models have achieved impressive performance on various downstream tasks. However, their large model sizes hinder their utilization on platforms with limited computational resources. We find that directly using smaller pre-trained models and applying magnitude-based pruning on CLIP models leads to inflexibility and inferior performance. Recent efforts for VLP compression either adopt uni-modal compression metrics resulting in limited performance or involve costly mask-search processes with learnable masks. In this paper, we first propose the Module-wise Pruning Error (MoPE) metric, accurately assessing CLIP module importance by performance decline on cross-modal tasks. Using the MoPE metric, we introduce a unified pruning framework applicable to both pre-training and task-specific fine-tuning compression stages. For pre-training, MoPE-CLIP effectively leverages knowledge from the teacher model, significantly reducing pre-training costs while maintaining strong zero-shot capabilities. For fine-tuning, consecutive pruning from width to depth yields highly competitive task-specific models. Extensive experiments in two stages demonstrate the effectiveness of the MoPE metric, and MoPE-CLIP outperforms previous state-of-the-art VLP compression methods.

Zhiming Mao, Haoli Bai, Lu Hou et al.

Prior study shows that pre-training techniques can boost the performance of visual document understanding (VDU), which typically requires models to gain abilities to perceive and reason both document texts and layouts (e.g., locations of texts and table-cells). To this end, we propose visually guided generative text-layout pre-training, named ViTLP. Given a document image, the model optimizes hierarchical language and layout modeling objectives to generate the interleaved text and layout sequence. In addition, to address the limitation of processing long documents by Transformers, we introduce a straightforward yet effective multi-segment generative pre-training scheme, facilitating ViTLP to process word-intensive documents of any length. ViTLP can function as a native OCR model to localize and recognize texts of document images. Besides, ViTLP can be effectively applied to various downstream VDU tasks. Extensive experiments show that ViTLP achieves competitive performance over existing baselines on benchmark VDU tasks, including information extraction, document classification, and document question answering.

Jian Yuan, Ziwei He, Haoli Bai et al.

Large Language Models (LLMs) use key-value (KV) cache to reduce redundant computation in autoregressive generation. However, the KV cache size increases linearly during generation, leading to excessive memory usage, especially for long texts. Most KV cache compression methods evict the unimportant KV pairs to maintain a fixed cache size, which leads to the permanent loss of tokens during generation. However, singular value decomposition shows that \textit{values} do not exhibit a strong low-rank property as \textit{keys} do, suggesting that information is distributed more evenly across \textit{values}, in contrast to its more redundant distribution within \textit{keys}. Therefore, methods that evict both \textit{keys} and \textit{values} risk losing crucial information and compromise context integrity, ultimately degrading the output quality. To address this problem, we propose WeightedKV, a novel, training-free approach that discards the \textit{keys} of less important tokens, while merging their \textit{values} into neighboring tokens via a convex combination weighted by their average attention scores. In this way, the retained \textit{keys} serve as anchors that guide the generation process, while the merged \textit{values} provide a rich contextual backdrop. We assess our method on four widely used language modeling datasets, demonstrating superior performance compared to all baseline methods, particularly with a lower budget ratio.

Ziwei He, Jian Yuan, Haoli Bai et al.

Efficient key-value (KV) cache compression is critical for scaling transformer-based Large Language Models (LLMs) in long sequences and resource-limited settings. Existing methods evict tokens based on their positions or importance scores, but position-based strategies can miss crucial information outside predefined regions, while those relying on global importance scores resulting in strong regional biases, limiting the KV cache's overall context retention and potentially impairing the performance of LLMs on complex tasks. Our wavelet analysis reveals that as tokens approach the end of sequence, their contributions to generation gradually increase and tends to diverge more from neighboring tokens, indicating a smooth transition with increasing complexity and variability from distant to nearby context. Motivated by this observation, we propose TreeKV, an intuitive, training-free method that employs a tree structure for smooth cache compression. TreeKV maintains a fixed cache size, allowing LLMs to deliver high-quality output even in long text scenarios. Unlike most compression methods, TreeKV is applicable to both the generation and prefilling stages. TreeKV consistently surpasses all baseline models in language modeling tasks on PG19 and OpenWebText2, allowing LLMs trained with short context window to generalize to longer window with a 16x cache reduction. On the Longbench benchmark, TreeKV achieves the best performance with only 6\% of the budget at optimal efficiency.

Tao Yuan, Haoli Bai, Yinfei Pan et al.

With the increasing size of large language models, layer pruning has gained increased attention as a hardware-friendly approach for model compression. However, existing layer pruning methods struggle to simultaneously address key practical deployment challenges, including performance degradation, high training costs, and limited acceleration. To overcome these limitations, we propose \name, a task-\underline{E}ffective, training-\underline{E}conomical and inference-\underline{E}fficient layer pruning framework. \namespace introduces two key innovations: (1) a differentiable mask optimization method using a Gumbel-TopK sampler, enabling efficient and precise pruning mask search; and (2) an entropy-aware adaptive knowledge distillation strategy that enhances task performance. Extensive experiments over diverse model architectures and benchmarks demonstrate the superiority of our method over state-of-the-art approaches. Notably, \namespace achieves 96\% accuracy, a mere 0.8\% drop from the original model (96.8\%) on MATH-500 when pruning 25\% layers of Qwen3-32B, outperforming existing SOTA (95\%), with a 1.33$\times$ inference speedup by consuming merely 0.5B tokens (0.5\% of the post-training data volume).

Huanyu Wang, Jushi Kai, Haoli Bai et al.

Vision-Language Models (VLMs) typically replace the predefined image placeholder token (<image>) in textual instructions with visual features from an image encoder, forming the input to a backbone Large Language Model (LLM). However, the large number of vision tokens significantly increases the context length, leading to high computational overhead and inference latency. While previous efforts mitigate this by selecting only important visual features or leveraging learnable queries to reduce token count, they often compromise performance or introduce substantial extra costs. In response, we propose Fourier-VLM, a simple yet efficient method that compresses visual representations in the frequency domain. Our approach is motivated by the observation that vision features output from the vision encoder exhibit concentrated energy in low-frequency components. Leveraging this, we apply a low-pass filter to the vision features using a two-dimensional Discrete Cosine Transform (DCT). Notably, the DCT is efficiently computed via the Fast Fourier Transform (FFT) operator with a time complexity of $\mathcal{O}(n\log n)$, minimizing the extra computational cost while introducing no additional parameters. Extensive experiments across various image-based benchmarks demonstrate that Fourier-VLM achieves competitive performance with strong generalizability across both LLaVA and Qwen-VL architectures. Crucially, it reduce inference FLOPs by up to 83.8% and boots generation speed by 31.2% compared to LLaVA-v1.5, highlighting the superior efficiency and practicality.

Jierun Chen, Tiezheng Yu, Haoli Bai et al.

Large vision-language models (VLMs) increasingly adopt post-training techniques such as long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL) to elicit sophisticated reasoning. While these methods exhibit synergy in language-only models, their joint effectiveness in VLMs remains uncertain. We present a systematic investigation into the distinct roles and interplay of long-CoT SFT and RL across multiple multimodal reasoning benchmarks. We find that SFT improves performance on difficult questions by in-depth, structured reasoning, but introduces verbosity and degrades performance on simpler ones. In contrast, RL promotes generalization and brevity, yielding consistent improvements across all difficulty levels, though the improvements on the hardest questions are less prominent compared to SFT. Surprisingly, combining them through two-staged, interleaved, or progressive training strategies, as well as data mixing and model merging, all fails to produce additive benefits, instead leading to trade-offs in accuracy, reasoning style, and response length. This ``synergy dilemma'' highlights the need for more seamless and adaptive approaches to unlock the full potential of combined post-training techniques for reasoning VLMs.

Zili Wang, Tianyu Zhang, Haoli Bai et al.

Test-Time Scaling (TTS) has proven effective in improving the performance of Large Language Models (LLMs) during inference. However, existing research has overlooked the efficiency of TTS from a latency-sensitive perspective. Through a latency-aware evaluation of representative TTS methods, we demonstrate that a compute-optimal TTS does not always result in the lowest latency in scenarios where latency is critical. To address this gap and achieve latency-optimal TTS, we propose two key approaches by optimizing the concurrency configurations: (1) branch-wise parallelism, which leverages multiple concurrent inference branches, and (2) sequence-wise parallelism, enabled by speculative decoding. By integrating these two approaches and allocating computational resources properly to each, our latency-optimal TTS enables a 32B model to reach 82.3% accuracy on MATH-500 within 1 minute and a smaller 3B model to achieve 72.4% within 10 seconds. Our work emphasizes the importance of latency-aware TTS and demonstrates its ability to deliver both speed and accuracy in latency-sensitive scenarios.

Haoli Bai, Hongda Mao, Dinesh Nair

As one of the successful Transformer-based models in computer vision tasks, SegFormer demonstrates superior performance in semantic segmentation. Nevertheless, the high computational cost greatly challenges the deployment of SegFormer on edge devices. In this paper, we seek to design a lightweight SegFormer for efficient semantic segmentation. Based on the observation that neurons in SegFormer layers exhibit large variances across different images, we propose a dynamic gated linear layer, which prunes the most uninformative set of neurons based on the input instance. To improve the dynamically pruned SegFormer, we also introduce two-stage knowledge distillation to transfer the knowledge within the original teacher to the pruned student network. Experimental results show that our method can significantly reduce the computation overhead of SegFormer without an apparent performance drop. For instance, we can achieve 36.9% mIoU with only 3.3G FLOPs on ADE20K, saving more than 60% computation with the drop of only 0.5% in mIoU

Haoli Bai, Lu Hou, Lifeng Shang et al.

Network quantization has gained increasing attention with the rapid growth of large pre-trained language models~(PLMs). However, most existing quantization methods for PLMs follow quantization-aware training~(QAT) that requires end-to-end training with full access to the entire dataset. Therefore, they suffer from slow training, large memory overhead, and data security issues. In this paper, we study post-training quantization~(PTQ) of PLMs, and propose module-wise quantization error minimization~(MREM), an efficient solution to mitigate these issues. By partitioning the PLM into multiple modules, we minimize the reconstruction error incurred by quantization for each module. In addition, we design a new model parallel training strategy such that each module can be trained locally on separate computing devices without waiting for preceding modules, which brings nearly the theoretical training speed-up (e.g., $4\times$ on $4$ GPUs). Experiments on GLUE and SQuAD benchmarks show that our proposed PTQ solution not only performs close to QAT, but also enjoys significant reductions in training time, memory overhead, and data consumption.

Xianghong Fang, Haoli Bai, Jian Li et al.

Variational autoencoders (VAEs) have been widely applied for text modeling. In practice, however, they are troubled by two challenges: information underrepresentation and posterior collapse. The former arises as only the last hidden state of LSTM encoder is transformed into the latent space, which is generally insufficient to summarize the data. The latter is a long-standing problem during the training of VAEs as the optimization is trapped to a disastrous local optimum. In this paper, we propose Discrete Auto-regressive Variational Attention Model (DAVAM) to address the challenges. Specifically, we introduce an auto-regressive variational attention approach to enrich the latent space by effectively capturing the semantic dependency from the input. We further design discrete latent space for the variational attention and mathematically show that our model is free from posterior collapse. Extensive experiments on language modeling tasks demonstrate the superiority of DAVAM against several VAE counterparts.

Haoli Bai, Wei Zhang, Lu Hou et al.

The rapid development of large pre-trained language models has greatly increased the demand for model compression techniques, among which quantization is a popular solution. In this paper, we propose BinaryBERT, which pushes BERT quantization to the limit by weight binarization. We find that a binary BERT is hard to be trained directly than a ternary counterpart due to its complex and irregular loss landscape. Therefore, we propose ternary weight splitting, which initializes BinaryBERT by equivalently splitting from a half-sized ternary network. The binary model thus inherits the good performance of the ternary one, and can be further enhanced by fine-tuning the new architecture after splitting. Empirical results show that our BinaryBERT has only a slight performance drop compared with the full-precision model while being 24x smaller, achieving the state-of-the-art compression results on the GLUE and SQuAD benchmarks.

Xianghong Fang, Haoli Bai, Zenglin Xu et al.

Variational autoencoders have been widely applied for natural language generation, however, there are two long-standing problems: information under-representation and posterior collapse. The former arises from the fact that only the last hidden state from the encoder is transformed to the latent space, which is insufficient to summarize data. The latter comes as a result of the imbalanced scale between the reconstruction loss and the KL divergence in the objective function. To tackle these issues, in this paper we propose the discrete variational attention model with categorical distribution over the attention mechanism owing to the discrete nature in languages. Our approach is combined with an auto-regressive prior to capture the sequential dependency from observations, which can enhance the latent space for language generation. Moreover, thanks to the property of discreteness, the training of our proposed approach does not suffer from posterior collapse. Furthermore, we carefully analyze the superiority of discrete latent space over the continuous space with the common Gaussian distribution. Extensive experiments on language generation demonstrate superior advantages of our proposed approach in comparison with the state-of-the-art counterparts.

Yuhang Li, Wei Wang, Haoli Bai et al.

Network quantization has rapidly become one of the most widely used methods to compress and accelerate deep neural networks. Recent efforts propose to quantize weights and activations from different layers with different precision to improve the overall performance. However, it is challenging to find the optimal bitwidth (i.e., precision) for weights and activations of each layer efficiently. Meanwhile, it is yet unclear how to perform convolution for weights and activations of different precision efficiently on generic hardware platforms. To resolve these two issues, in this paper, we first propose an Efficient Bitwidth Search (EBS) algorithm, which reuses the meta weights for different quantization bitwidth and thus the strength for each candidate precision can be optimized directly w.r.t the objective without superfluous copies, reducing both the memory and computational cost significantly. Second, we propose a binary decomposition algorithm that converts weights and activations of different precision into binary matrices to make the mixed precision convolution efficient and practical. Experiment results on CIFAR10 and ImageNet datasets demonstrate our mixed precision QNN outperforms the handcrafted uniform bitwidth counterparts and other mixed precision techniques.

Yuhang Li, Xin Dong, Sai Qian Zhang et al.

To deploy deep neural networks on resource-limited devices, quantization has been widely explored. In this work, we study the extremely low-bit networks which have tremendous speed-up, memory saving with quantized activation and weights. We first bring up three omitted issues in extremely low-bit networks: the squashing range of quantized values; the gradient vanishing during backpropagation and the unexploited hardware acceleration of ternary networks. By reparameterizing quantized activation and weights vector with full precision scale and offset for fixed ternary vector, we decouple the range and magnitude from the direction to extenuate the three issues. Learnable scale and offset can automatically adjust the range of quantized values and sparsity without gradient vanishing. A novel encoding and computation pat-tern are designed to support efficient computing for our reparameterized ternary network (RTN). Experiments on ResNet-18 for ImageNet demonstrate that the proposed RTN finds a much better efficiency between bitwidth and accuracy, and achieves up to 26.76% relative accuracy improvement compared with state-of-the-art methods. Moreover, we validate the proposed computation pattern on Field Programmable Gate Arrays (FPGA), and it brings 46.46x and 89.17x savings on power and area respectively compared with the full precision convolution.

Haoli Bai, Jiaxiang Wu, Irwin King et al.

Model compression has been widely adopted to obtain light-weighted deep neural networks. Most prevalent methods, however, require fine-tuning with sufficient training data to ensure accuracy, which could be challenged by privacy and security issues. As a compromise between privacy and performance, in this paper we investigate few shot network compression: given few samples per class, how can we effectively compress the network with negligible performance drop? The core challenge of few shot network compression lies in high estimation errors from the original network during inference, since the compressed network can easily over-fits on the few training instances. The estimation errors could propagate and accumulate layer-wisely and finally deteriorate the network output. To address the problem, we propose cross distillation, a novel layer-wise knowledge distillation approach. By interweaving hidden layers of teacher and student network, layer-wisely accumulated estimation errors can be effectively reduced.The proposed method offers a general framework compatible with prevalent network compression techniques such as pruning. Extensive experiments on benchmark datasets demonstrate that cross distillation can significantly improve the student network's accuracy when only a few training instances are available.

Liangjian Wen, Xuanyang Zhang, Haoli Bai et al.

Recurrent neural networks (RNNs) have recently achieved remarkable successes in a number of applications. However, the huge sizes and computational burden of these models make it difficult for their deployment on edge devices. A practically effective approach is to reduce the overall storage and computation costs of RNNs by network pruning techniques. Despite their successful applications, those pruning methods based on Lasso either produce irregular sparse patterns in weight matrices, which is not helpful in practical speedup. To address these issues, we propose structured pruning method through neuron selection which can reduce the sizes of basic structures of RNNs. More specifically, we introduce two sets of binary random variables, which can be interpreted as gates or switches to the input neurons and the hidden neurons, respectively. We demonstrate that the corresponding optimization problem can be addressed by minimizing the L0 norm of the weight matrix. Finally, experimental results on language modeling and machine reading comprehension tasks have indicated the advantages of the proposed method in comparison with state-of-the-art pruning competitors. In particular, nearly 20 x practical speedup during inference was achieved without losing performance for language model on the Penn TreeBank dataset, indicating the promising performance of the proposed method

Xianghong Fang, Haoli Bai, Ziyi Guo et al.

The accuracy of deep learning (e.g., convolutional neural networks) for an image classification task critically relies on the amount of labeled training data. Aiming to solve an image classification task on a new domain that lacks labeled data but gains access to cheaply available unlabeled data, unsupervised domain adaptation is a promising technique to boost the performance without incurring extra labeling cost, by assuming images from different domains share some invariant characteristics. In this paper, we propose a new unsupervised domain adaptation method named Domain-Adversarial Residual-Transfer (DART) learning of Deep Neural Networks to tackle cross-domain image classification tasks. In contrast to the existing unsupervised domain adaption approaches, the proposed DART not only learns domain-invariant features via adversarial training, but also achieves robust domain-adaptive classification via a residual-transfer strategy, all in an end-to-end training framework. We evaluate the performance of the proposed method for cross-domain image classification tasks on several well-known benchmark data sets, in which our method clearly outperforms the state-of-the-art approaches.

Hao Liu, Haoli Bai, Lirong He et al.

Unsupervised structure learning in high-dimensional time series data has attracted a lot of research interests. For example, segmenting and labelling high dimensional time series can be helpful in behavior understanding and medical diagnosis. Recent advances in generative sequential modeling have suggested to combine recurrent neural networks with state space models (e.g., Hidden Markov Models). This combination can model not only the long term dependency in sequential data, but also the uncertainty included in the hidden states. Inheriting these advantages of stochastic neural sequential models, we propose a structured and stochastic sequential neural network, which models both the long-term dependencies via recurrent neural networks and the uncertainty in the segmentation and labels via discrete random variables. For accurate and efficient inference, we present a bi-directional inference network by reparamterizing the categorical segmentation and labels with the recent proposed Gumbel-Softmax approximation and resort to the Stochastic Gradient Variational Bayes. We evaluate the proposed model in a number of tasks, including speech modeling, automatic segmentation and labeling in behavior understanding, and sequential multi-objects recognition. Experimental results have demonstrated that our proposed model can achieve significant improvement over the state-of-the-art methods.