As AI assistance becomes embedded in programming practice, researchers have increasingly examined how these systems help learners generate code and work more efficiently. However, these studies often position AI as a replacement for human collaboration and overlook the social and learning-oriented aspects that emerge in collaborative programming. Our work introduces human-human-AI (HHAI) triadic programming, where an AI agent serves as an additional collaborator rather than a substitute for a human partner. Through a within-subjects study with 20 participants, we show that triadic collaboration enhances collaborative learning and social presence compared to the dyadic human–AI (HAI) baseline. In the triadic HHAI conditions, participants relied significantly less on AI generated code in their work. This effect was strongest in the HHAI-shared condition, where participants had an increased sense of responsibility to understand AI suggestions before applying them. These findings demonstrate how triadic settings activate socially shared regulation of learning by making AI use visible and accountable to a human peer, suggesting that AI systems that augment rather than automate peer collaboration can better preserve the learning processes that collaborative programming relies on.

Autistic individuals sometimes disclose autism when asking LLMs for social advice, hoping for more personalized responses. However, they also recognize that these systems may reproduce stereotypes, raising uncertainty about the risks and benefits of disclosure. We conducted a mixed-methods study combining a large-scale LLM audit experiment with interviews involving 11 autistic participants. We developed a six-step pipeline operationalizing 12 documented autism stereotypes into decision-making scenarios framed as users requesting advice (e.g., “Should I do A or B?”). We generated 345,000 responses from six LLMs and measured how advice shifted when prompts disclosed autism versus when they did not. When autism was disclosed, LLMs disproportionately recommended avoiding stereotypically stressful situations, including social events, confrontations, new experiences, and romantic relationships. While some participants viewed this as affirming, others criticized it as infantilizing or undermining opportunities for growth. Our study illuminates how the intermingling of affirmation and stereotyping complicates the personalization of LLMs.1

Autistic individuals often experience negative self-talk (NST), leading to increased anxiety and depression. While therapy is recommended, it presents challenges for many autistic individuals. Meanwhile, a growing number are turning to large language models (LLMs) for mental health support. To understand how autistic individuals perceive AI’s role in coping with NST, we surveyed 200 autistic adults and interviewed practitioners. We also analyzed LLM responses to participants’ hypothetical prompts about their NST. Our findings show that participants view LLMs as useful for managing NST by identifying and reframing negative thoughts. Both participants and practitioners recognize AI’s potential to support therapy and emotional expression. Participants also expressed concerns about LLMs’ understanding of neurodivergent thought patterns, particularly due to the neurotypical bias of LLMs. Practitioners critiqued LLMs’ responses as overly wordy, vague, and overwhelming. This study contributes to the growing research on AI-assisted mental health support, with specific insights for supporting the autistic community.

Understanding causal language in informal discourse is a core yet underexplored challenge in NLP.Existing datasets largely focus on explicit causality in structured text, providing limited support for detecting implicit causal expressions, particularly those found in informal, user-generated social media posts.We introduce CausalTalk, a multi-level dataset of five years of Reddit posts (2020-2024) discussing public health related to the COVID-19 pandemic, among which 10, 120 posts are annotated across four causal tasks: (1) binary causal classification, (2) explicit vs. implicit causality, (3) cause-effect span extraction, and (4) causal gist generation.Annotations comprise both gold-standard labels created by domain experts and silver-standard labels generated by GPT-4o and verified by human annotators.CausalTalk bridges fine-grained causal detection and gistbased reasoning over informal text.It enables benchmarking across both discriminative and generative models, and provides a rich resource for studying causal reasoning in social media contexts 1 .

One challenge in technical interviews is the thinkaloud process, where candidates verbalize their thought processes while solving coding tasks. Despite its importance, opportunities for structured practice remain limited. Conversational AI offers potential assistance, but limited research explores user perceptions of its role in think-aloud practice. To address this gap, we conducted a study with 17 participants using an LLM-based technical interview practice tool. Participants valued AI’s role in simulation, feedback, and learning from generated examples. Key design recommendations include promoting social presence in conversational AI for technical interview simulation, providing feedback beyond verbal content analysis, and enabling crowdsourced think-aloud examples through humanAI collaboration. Beyond feature design, we examined broader considerations, including intersectional challenges and potential strategies to address them, how AI-driven interview preparation could promote equitable learning in computing careers, and the need to rethink AI’s role in interview practice by suggesting a research direction that integrates human-AI collaboration.

Online hate speech has become increasingly prevalent on social media, causing harm to individuals and society. While automated content moderation has received considerable attention, user-driven counterspeech remains a less explored yet promising approach. However, many people face difficulties in crafting effective responses. We introduce CounterQuill, a human-AI collaborative system that helps everyday users with writing empathetic counterspeech-not by generating automatic replies, but by educating them through reflection and response. CounterQuill follows a three-stage workflow grounded in computational thinking: (1) a learning session to build understanding of hate speech and counterspeech, (2) a brainstorming session to identify harmful patterns and ideate counterspeech ideas, and (3) a co-writing session that helps users refine their counter responses while preserving personal voice. Through a user study ($\mathbf{N} \boldsymbol{=} \mathbf{2 0}$), we found that CounterQuill helped participants develop the skills to brainstorm and draft counterspeech with confidence and control throughout the process. Our findings highlight how AI systems can scaffold complex communication tasks through structured, human-centered workflows that educate users on how to recognize, reflect on, and respond to online hate speech.

Despite the increasing use of large language models (LLMs) in everyday life among neurodivergent individuals, our knowledge of how they engage with and perceive LLMs remains limited. In this study, we investigate how neurodivergent individuals interact with LLMs by qualitatively analyzing topically related discussions from 61 neurodivergent communities on Reddit. Our findings reveal 20 specific LLM use cases across five core thematic areas of use among neurodivergent users: emotional well-being, mental health support, interpersonal communication, learning, and professional development and productivity. We also identified key challenges, including overly neurotypical LLM responses and the limitations of text-based interactions. In response to such challenges, some users actively seek advice by sharing input prompts and corresponding LLM responses. Others develop workarounds by experimenting and modifying prompts to be more neurodivergent-friendly. Despite these efforts, users have significant concerns around LLM use, including potential overreliance and fear of replacing human connections. Our analysis highlights the need to make LLMs more inclusive for neurodivergent users and implications around how LLM technologies can reinforce unintended consequences and behaviors.

Job interviews play a critical role in shaping one's career, yet practicing interview skills can be challenging, especially without access to human coaches or peers for feedback. Recent advancements in large language models (LLMs) present an opportunity to enhance the interview practice experience. Yet, little research has explored the effectiveness and user perceptions of such systems or the benefits and challenges of using LLMs for interview practice. Furthermore, while prior work and recent commercial tools have demonstrated the potential of AI to assist with interview practice, they often deliver one-way feedback, where users only receive information about their performance. By contrast, dialogic feedback , a concept developed in learning sciences, is a two-way interaction feedback process that allows users to further engage with and learn from the provided feedback through interactive dialogue. This paper introduces Conversate, a web-based application that supports reflective learning in job interview practice by leveraging large language models (LLMs) for interactive interview simulations and dialogic feedback. To start the interview session, the user provides the title of a job position (e.g., entry-level software engineer) in the system. Then, our system will initialize the LLM agent to start the interview simulation by asking the user an opening interview question and following up with questions carefully adapted to subsequent user responses. After the interview session, our back-end LLM framework will then analyze the user's responses and highlight areas for improvement. Users can then annotate the transcript by selecting specific sections and writing self-reflections. Finally, the user can interact with the system for dialogic feedback, conversing with the LLM agent to learn from and iteratively refine their answers based on the agent's guidance. To evaluate Conversate, we conducted a user study with 19 participants to understand their perceptions of using LLM-supported interview simulation and dialogic feedback. Our findings show that participants valued the adaptive follow-up questions from LLMs, as they enhanced the realism of interview simulations and encouraged deeper thinking. Participants also appreciated the AI-assisted annotation, as it reduced their cognitive burden and mitigated excessive self-criticism in their own evaluation of their interview performance. Moreover, participants found the LLM-supported dialogic feedback to be beneficial, as it promoted personalized and continuous learning, reduced feelings of judgment, and allowed them to express disagreement.

This paper proposes the paradigm of large convolutional kernels in designing modern Convolutional Neural Networks (ConvNets). We establish that employing a few large kernels, instead of stacking multiple smaller ones, can be a superior design strategy. Our work introduces a set of architecture design guidelines for large-kernel ConvNets that optimize their efficiency and performance. We propose the UniRepLKNet architecture, which offers systematical architecture design principles specifically crafted for large-kernel ConvNets, emphasizing their unique ability to capture extensive spatial information without deep layer stacking. This results in a model that not only surpasses its predecessors with an ImageNet accuracy of 88.0%, an ADE20 K mIoU of 55.6%, and a COCO box AP of 56.4% but also demonstrates impressive scalability and performance on various modalities such as time-series forecasting, audio, point cloud, and video recognition. These results indicate the universal modeling abilities of large-kernel ConvNets with faster inference speed compared with vision transformers. Our findings reveal that large-kernel ConvNets possess larger effective receptive fields and a higher shape bias, moving away from the texture bias typical of smaller-kernel CNNs.

We propose reparameterized refocusing convolution (RefConv) as a replacement for regular convolutional layers, which is a plug-and-play module to improve the performance without any inference costs. Specifically, given a pretrained model, RefConv applies a trainable Refocusing Transformation to the basis kernels inherited from the pretrained model to establish connections among the parameters. For example, a depthwise RefConv can relate the parameters of a specific channel of convolution kernel to the parameters of the other kernel, i.e., make them refocus on the other parts of the model they have never attended to, rather than focus on the input features only. From another perspective, RefConv augments the priors of existing model structures by utilizing the representations encoded in the pretrained parameters as the priors and refocusing on them to learn novel representations, thus further enhancing the representational capacity of the pretrained model. The experimental results validated that RefConv can improve multiple convolutional neural network (CNN)-based models by a clear margin on image classification (up to 1.47% higher top-1 accuracy on ImageNet), object detection, semantic segmentation, and adversarial attacks without introducing any extra inference costs or altering the original model structure. Further studies demonstrated that RefConv can strengthen the spatial skeletons of kernels, reduce the redundancy of channels, and smooth the loss landscape, which explains its effectiveness.

Current research mainly explores the attributes and impact of online counterspeech, leaving a gap in understanding of who engages in online counterspeech or what motivates or deters users from participating. To investigate this, we surveyed 458 English-speaking US participants, analyzing key motivations and barriers underlying online counterspeech engagement. We presented each participant with three hate speech examples from a set of 900, spanning race, gender, religion, sexual orientation, and disability and requested counterspeech responses. Subsequent questions assessed their satisfaction, perceived difficulty, and the effectiveness of their counterspeech. Our findings show that having been a target of online hate is a key driver of frequent online counterspeech engagement. People differ in their motivations and barriers toward engaging in online counterspeech across different demographic groups. Younger individuals, women, those with higher education levels, and regular witnesses to online hate are more reluctant to engage in online counterspeech due to concerns around public exposure, retaliation, and third-party harassment. Varying motivation and barriers in counterspeech engagement also shape how individuals view their own self-authored counterspeech and the difficulty experienced writing it. Additionally, our work explores people’s willingness to use AI technologies like ChatGPT for counterspeech writing. Through this work we introduce a multi-item scale for understanding counterspeech motivation and barriers and a more nuanced understanding of the factors shaping online counterspeech engagement.

We introduce a multi-step reasoning framework using prompt-based LLMs to examine the relationship between social media language patterns and trends in national health outcomes. Grounded in fuzzy-trace theory, which emphasizes the importance of “gists” of causal coherence in effective health communication, we introduce Role-Based Incremental Coaching (RBIC), a prompt-based LLM framework, to identify gists at-scale. Using RBIC, we systematically extract gists from subreddit discussions opposing COVID-19 health measures (Study 1). We then track how these gists evolve across key events (Study 2) and assess their influence on online engagement (Study 3). Finally, we investigate how the volume of gists is associated with national health trends like vaccine uptake and hospitalizations (Study 4). Our work is the first to empirically link social media linguistic patterns to real-world public health trends, highlighting the potential of prompt-based LLMs in identifying critical online discussion patterns that can form the basis of public health communication strategies.

Large-kernel convolutional neural networks (ConvNets) have recently received extensive research attention, but two unresolved and critical issues demand further investigation. 1) The architectures of existing large-kernel ConvNets largely follow the design principles of conventional ConvNets or transformers, while the architectural design for large-kernel ConvNets remains under-addressed. 2) As transformers have dominated multiple modalities, it re-mains to be investigated whether ConvNets also have a strong universal perception ability in domains beyond vision. In this paper, we contribute from two aspects. 1) We propose four architectural guidelines for designing large- kernel ConvNets, the core of which is to exploit the essential characteristics of large kernels that distinguish them from small kernels - they can see wide without going deep. Fol-lowing such guidelines, our proposed large-kernel ConvNet shows leading performance in image recognition (ImageNet accuracy of 88.0%, ADE20K mIoU of 55.6%, and COCO box AP of 56.4%), demonstrating better performance and higher speed than the recent powerful competitors. 2) We discover large kernels are the key to unlocking the exceptional performance of ConvNets in domains where they were originally not proficient. With certain modality-related pre-processing approaches, the proposed model achieves state- of-the-art performance on time-series forecasting and audio recognition tasks even without modality-specific customization to the architecture. All the code and models are publicly available on GitHub and Huggingface.

This paper focuses on the high computational complexity in Large Language Models (LLMs), a significant challenge in both natural language processing (NLP) and multi-modal tasks. We propose Low-Rank Approximation for Sparse Attention (LoRA -Sparse), an innovative approach that strategically reduces this complexity. LoRA -Sparse introduces low-rank linear projection layers for sparse attention approximation. It utilizes an order-mimic training methodology, which is crucial for efficiently approximating the self-attention mechanism in LLMs. We empirically show that sparse attention not only reduces computational demands, but also enhances model performance in both NLP and multi-modal tasks. This surprisingly shows that redundant attention in LLMs might be non-beneficial. We extensively validate LoRA -Sparse through rigorous empirical studies in both (NLP) and multi-modal tasks, demonstrating its effectiveness and general applicability. Based on LLaMA and LLaVA models, our methods can reduce more than half of the self-attention computation with even better performance than full-attention baselines.

We propose to improve transformers of a specific modality with irrelevant data from other modalities, e.g., improve an ImageNet model with audio or point cloud datasets. We would like to highlight that the data samples of the target modality are irrelevant to the other modalities, which distinguishes our method from other works utilizing paired (e.g., CLIP) or interleaved data of different modalities. We propose a methodology named Multimodal Pathway - given a target modality and a transformer designed for it, we use an auxiliary transformer trained with data of another modality and construct pathways to connect components of the two models so that data of the target modality can be processed by both models. In this way, we utilize the universal sequence-to-sequence modeling abilities of transformers obtained from two modalities. As a concrete implementation, we use a modality-specific tokenizer and task-specific head as usual but utilize the transformer blocks of the auxiliary model via a proposed method named Cross-Modal Re-parameterization, which exploits the auxiliary weights without any inference costs. On the image, point cloud, video, and audio recognition tasks, we observe significant and consistent performance improvements with irrelevant data from other modalities. The code and models are available at https://github.com/AILab-CVC/M2PT.

In this work, we examine the linguistic signature of online racial microaggressions (acts) and how it differs from that of personal narratives recalling experiences of such aggressions (recalls) by Black social media users. We manually curate and annotate a corpus of acts and recalls fromin-the-wild social media discussions, and verify labels with Black workshop participants. We leverage Natural Language Processing (NLP) and qualitative analysis on this data to classify (RQ1), interpret (RQ2), and characterize (RQ3) the language underlying acts and recalls of racial microaggressions in the context of racism in the U.S. Our findings show that neural language models (LMs) can classify acts and recalls with high accuracy (RQ1) with contextual words revealing themes that associate Blacks with objects that reify negative stereotypes (RQ2). Furthermore, overlapping linguistic signatures between acts and recalls serve functionally different purposes (RQ3), providing broader implications to the current challenges in content moderation systems on social media.

The existence of redundancy in convolutional neural networks (CNNs) enables us to remove some filters/channels with acceptable performance drops. However, the training objective of CNNs usually tends to minimize an accuracy-related loss function without any attention paid to the redundancy, making the redundancy distribute randomly on all the filters, such that removing any of them may trigger information loss and accuracy drop, necessitating a fine-tuning step for recovery. In this article, we propose to manipulate the redundancy during training to facilitate network pruning. To this end, we propose a novel centripetal SGD (C-SGD) to make some filters identical, resulting in ideal redundancy patterns, as such filters become purely redundant due to their duplicates, hence removing them does not harm the network. As shown on CIFAR and ImageNet, C-SGD delivers better performance because the redundancy is better organized, compared to the existing methods. The efficiency also characterizes C-SGD because it is as fast as regular SGD, requires no fine-tuning, and can be conducted simultaneously on all the layers even in very deep CNNs. Besides, C-SGD can improve the accuracy of CNNs by first training a model with the same architecture but wider layers and then squeezing it into the original width.

In zero-shot learning (ZSL), generative methods synthesize class-related sample features based on predefined semantic prototypes. They advance the ZSL performance by synthesizing unseen class sample features for better training the classifier. We observe that each class’s predefined semantic prototype (also referred to as semantic embedding or condition) does not accurately match its real semantic prototype. So the synthesized visual sample features do not faithfully represent the real sample features, limiting the classifier training and existing ZSL performance. In this paper, we formulate this mismatch phenomenon as the visual-semantic domain shift problem. We propose a dynamic semantic prototype evolving (DSP) method to align the empirically predefined semantic prototypes and the real prototypes for class-related feature synthesis. The alignment is learned by refining sample features and semantic prototypes in a unified framework and making the synthesized visual sample features approach real sample features. After alignment, synthesized sample features from unseen classes are closer to the real sample features and benefit DSP to improve existing generative ZSL methods by 8.5%, 8.0%, and 9.7% on the standard CUB, SUN AWA2 datasets, the significant performance improvement indicates that evolving semantic prototype explores a virgin field in ZSL.

With the growth of online news over the past decade, empirical studies on political discourse and news consumption have focused on the phenomenon of filter bubbles and echo chambers. Yet recently, scholars have revealed limited evidence around the impact of such phenomenon, leading some to argue that partisan segregation across news audiences can- not be fully explained by online news consumption alone and that the role of traditional legacy media may be as salient in polarizing public discourse around current events. In this work, we expand the scope of analysis to include both online and more traditional media by investigating the relationship between broadcast news media language and social media discourse. By analyzing a decade’s worth of closed captions (2.1 million speaker turns) from CNN and Fox News along with topically corresponding discourse from Twitter, we pro- vide a novel framework for measuring semantic polarization between America’s two major broadcast networks to demonstrate how semantic polarization between these outlets has evolved (Study 1), peaked (Study 2) and influenced partisan discussions on Twitter (Study 3) across the last decade. Our results demonstrate a sharp increase in polarization in how topically important keywords are discussed between the two channels, especially after 2016, with overall highest peaks occurring in 2020. The two stations discuss identical topics in drastically distinct contexts in 2020, to the extent that there is barely any linguistic overlap in how identical keywords are contextually discussed. Further, we demonstrate at-scale, how such partisan division in broadcast media language significantly shapes semantic polarity trends on Twitter (and vice-versa), empirically linking for the first time, how online discussions are influenced by televised media. We show how the language characterizing opposing media narratives about similar news events on TV can increase levels of partisan dis- course online. To this end, our work has implications for how media polarization on TV plays a significant role in impeding rather than supporting online democratic discourse.

We revisit large kernel design in modern convolutional neural networks (CNNs). Inspired by recent advances in vision transformers (ViTs), in this paper, we demonstrate that using a few large convolutional kernels instead of a stack of small kernels could be a more powerful paradigm. We suggested five guidelines, e.g., applying re-parameterized large depthwise convolutions, to design efficient high-performance large-kernel CNNs. Following the guidelines, we propose RepLKNet, a pure CNN architecture whose kernel size is as large as 31×31, in contrast to commonly used 3×3. RepLKNet greatly closes the performance gap between CNNs and ViTs, e.g., achieving comparable or superior results than Swin Transformer on ImageNet and a few typical downstream tasks, with lower latency. RepLKNet also shows nice scalability to big data and large models, obtaining 87.8% top-1 accuracy on ImageNet and 56.0% mIoU on ADE20K, which is very competitive among the state-of-the-arts with similar model sizes. Our study further reveals that, in contrast to small-kernel CNNs, large-kernel CNNs have much larger effective receptive fields and higher shape bias rather than texture bias. Code & models at https://github.com/megvii-research/RepLKNet.

Compared to convolutional layers, fully-connected (FC) layers are better at modeling the long-range dependencies but worse at capturing the local patterns, hence usually less favored for image recognition. In this paper, we propose a methodology, Locality Injection, to incorporate local priors into an FC layer via merging the trained parameters of a parallel conv kernel into the FC kernel. Locality Injection can be viewed as a novel Structural Re-parameterization method since it equivalently converts the structures via transforming the parameters. Based on that, we propose a multi-layer-perceptron (MLP) block named RepMLP Block, which uses three FC layers to extract features, and a novel architecture named RepMLPNet. The hierarchical design distinguishes RepMLPNet from the other concurrently proposed vision MLPs. As it produces feature maps of different levels, it qualifies as a backbone model for downstream tasks like semantic segmentation. Our results reveal that 1) Locality Injection is a general methodology for MLP models; 2) RepMLPNet has favorable accuracy-efficiency trade-off compared to the other MLPs; 3) RepMLPNet is the first MLP that seamlessly transfer to Cityscapes semantic segmentation. The code and models are available at https://github.com/DingXiaoH/RepMLP.

End-to-end (E2E) automatic speech recognition models like Recurrent Neural Networks Transducer (RNN-T) are becoming a popular choice for streaming ASR applications like voice assistants. While E2E models are very effective at learning representation of the training data they are trained on, their accuracy on unseen domains remains a challenging problem. Additionally, these models require paired audio and text training data, are computationally expensive and are difficult to adapt towards the fast evolving nature of conversational speech. In this work, we explore a contextual biasing approach using likelihood-ratio that leverages text data sources to adapt RNN-T model to new domains and entities. We show that this method is effective in improving rare words recognition, and results in a relative improvement of 10% in 1-best word error rate (WER) and 10% in n-best Oracle1WER (n=8) on multiple out-of-domain datasets without any degradation on a general dataset. We also show that complementing the contextual biasing adaptation with adaptation of a second-pass rescoring model gives additive WER improvements.

We propose a universal building block of Convolutional Neural Network (ConvNet) to improve the performance without any inference-time costs. The block is named Diverse Branch Block (DBB), which enhances the representational capacity of a single convolution by combining diverse branches of different scales and complexities to enrich the feature space, including sequences of convolutions, multiscale convolutions, and average pooling. After training, a DBB can be equivalently converted into a single conv layer for deployment. Unlike the advancements of novel ConvNet architectures, DBB complicates the training-time microstructure while maintaining the macro architecture, so that it can be used as a drop-in replacement for regular conv layers of any architecture. In this way, the model can be trained to reach a higher level of performance and then transformed into the original inference-time structure for inference. DBB improves ConvNets on image classification (up to 1.9% higher top-1 accuracy on ImageNet), object detection and semantic segmentation. The PyTorch code and models are released at https://github.com/DingXiaoH/DiverseBranchBlock.

We present a simple but powerful architecture of convolutional neural network, which has a VGG-like inference-time body composed of nothing but a stack of 3 × 3 convolution and ReLU, while the training-time model has a multi-branch topology. Such decoupling of the training-time and inference-time architecture is realized by a structural re-parameterization technique so that the model is named RepVGG. On ImageNet, RepVGG reaches over 80% top-1 accuracy, which is the first time for a plain model, to the best of our knowledge. On NVIDIA 1080Ti GPU, RepVGG models run 83% faster than ResNet-50 or 101% faster than ResNet-101 with higher accuracy and show favorable accuracy-speed trade-off compared to the state-of-the-art models like EfficientNet and RegNet. The code and trained models are available at https://github.com/megvii-model/RepVGG.

We propose ResRep, a novel method for lossless channel pruning (a.k.a. filter pruning), which slims down a CNN by reducing the width (number of output channels) of convolutional layers. Inspired by the neurobiology research about the independence of remembering and forgetting, we propose to re-parameterize a CNN into the remembering parts and forgetting parts, where the former learn to maintain the performance and the latter learn to prune. Via training with regular SGD on the former but a novel update rule with penalty gradients on the latter, we realize structured sparsity. Then we equivalently merge the remembering and forgetting parts into the original architecture with narrower layers. In this sense, ResRep can be viewed as a successful application of Structural Re-parameterization. Such a methodology distinguishes ResRep from the traditional learning-based pruning paradigm that applies a penalty on parameters to produce sparsity, which may suppress the parameters essential for the remembering. ResRep slims down a standard ResNet-50 with 76.15% accuracy on ImageNet to a narrower one with only 45% FLOPs and no accuracy drop, which is the first to achieve lossless pruning with such a high compression ratio. The code and models are at https://github.com/DingXiaoH/ResRep.

Zero-shot learning (ZSL) is to build recognition models for previously unseen target classes which have no labeled data for training by transferring knowledge from some other related auxiliary source classes with abundant labeled samples to the target ones with class attributes as the bridge. The key is to learn a similarity based ranking function between samples and class labels using the labeled source classes so that the proper (unseen) class label for a test sample can be identified by the function. In order to learn the function, single-view ranking based loss is widely used which aims to rank the true label prior to the other labels for a training sample. However, we argue that the ranking can be performed from the other view, which aims to place the images belonging to a label before the images from the other classes. Motivated by it, we propose a novel DuAl-view RanKing (DARK) loss for zeroshot learning simultaneously ranking labels for an image by point-to-point metric and ranking images for a label by pointto-set metric, which is capable of better modeling the relationship between images and classes. In addition, we also notice that previous ZSL approaches mostly fail to well exploit the hardness of training samples, either using only very hard ones or using all samples indiscriminately. In this work, we also introduce a sample hardness assessment method to ZSL which assigns different weights to training samples based on their hardness, which leads to a more accurate and robust ZSL model. Experiments on benchmarks demonstrate that DARK outperforms the state-of-the-arts for (generalized) ZSL.

The redundancy is widely recognized in Convolutional Neural Networks (CNNs), which enables to remove some unimportant filters from convolutional layers so as to slim the network with acceptable performance drop. Inspired by the linearity of convolution, we seek to make some filters increasingly close and eventually identical for network slimming. To this end, we propose Centripetal SGD (C-SGD), a novel optimization method, which can train several filters to collapse into a single point in the parameter hyperspace. When the training is completed, the removal of the identical filters can trim the network with NO performance loss, thus no finetuning is needed. By doing so, we have partly solved an open problem of constrained filter pruning on CNNs with complicated structure, where some layers must be pruned following the others. Our experimental results on CIFAR-10 and ImageNet have justified the effectiveness of C-SGD-based filter pruning. Moreover, we have provided empirical evidences for the assumption that the redundancy in deep neural networks helps the convergence of training by showing that a redundant CNN trained using C-SGD outperforms a normally trained counterpart with the equivalent width.

As designing appropriate Convolutional Neural Network (CNN) architecture in the context of a given application usually involves heavy human works or numerous GPU hours, the research community is soliciting the architecture-neutral CNN structures, which can be easily plugged into multiple mature architectures to improve the performance on our real-world applications. We propose Asymmetric Convolution Block (ACB), an architecture-neutral structure as a CNN building block, which uses 1D asymmetric convolutions to strengthen the square convolution kernels. For an off-the-shelf architecture, we replace the standard square-kernel convolutional layers with ACBs to construct an Asymmetric Convolutional Network (ACNet), which can be trained to reach a higher level of accuracy. After training, we equivalently convert the ACNet into the same original architecture, thus requiring no extra computations anymore. We have observed that ACNet can improve the performance of various models on CIFAR and ImageNet by a clear margin. Through further experiments, we attribute the effectiveness of ACB to its capability of enhancing the model's robustness to rotational distortions and strengthening the central skeleton parts of square convolution kernels.

It is not easy to design and run Convolutional Neural Networks (CNNs) due to: 1) finding the optimal number of filters (i.e., the width) at each layer is tricky, given an architecture; and 2) the computational intensity of CNNs impedes the deployment on computationally limited devices. Oracle Pruning is designed to remove the unimportant filters from a well-trained CNN, which estimates the filters’ importance by ablating them in turn and evaluating the model, thus delivers high accuracy but suffers from intolerable time complexity, and requires a given resulting width but cannot automatically find it. To address these problems, we propose Approximated Oracle Filter Pruning (AOFP), which keeps searching for the least important filters in a binary search manner, makes pruning attempts by masking out filters randomly, accumulates the resulting errors, and finetunes the model via a multi-path framework. As AOFP enables simultaneous pruning on multiple layers, we can prune an existing very deep CNN with acceptable time cost, negligible accuracy drop, and no heuristic knowledge, or re-design a model which exerts higher accuracy and faster inference.

Deep Neural Network (DNN) is powerful but computationally expensive and memory intensive, thus impeding its practical usage on resource-constrained front-end devices. DNN pruning is an approach for deep model compression, which aims at eliminating some parameters with tolerable performance degradation. In this paper, we propose a novel momentum-SGD-based optimization method to reduce the network complexity by on-the-fly pruning. Concretely, given a global compression ratio, we categorize all the parameters into two parts at each training iteration which are updated using different rules. In this way, we gradually zero out the redundant parameters, as we update them using only the ordinary weight decay but no gradients derived from the objective function. As a departure from prior methods that require heavy human works to tune the layer-wise sparsity ratios, prune by solving complicated non-differentiable problems or finetune the model after pruning, our method is characterized by 1) global compression that automatically finds the appropriate per-layer sparsity ratios; 2) end-to-end training; 3) no need for a time-consuming re-training process after pruning; and 4) superior capability to find better winning tickets which have won the initialization lottery.

In recent years considerable research efforts have been devoted to compression techniques of convolutional neural networks (CNNs). Many works so far have focused on CNN connection pruning methods which produce sparse parameter tensors in convolutional or fully-connected layers. It has been demonstrated in several studies that even simple methods can effectively eliminate connections of a CNN. However, since these methods make parameter tensors just sparser but no smaller, the compression may not transfer directly to acceleration without support from specially designed hardware. In this paper, we propose an iterative approach named Auto-balanced Filter Pruning, where we pre-train the network in an innovative auto-balanced way to transfer the representational capacity of its convolutional layers to a fraction of the filters, prune the redundant ones, then re-train it to restore the accuracy. In this way, a smaller version of the original network is learned and the floating-point operations (FLOPs) are reduced. By applying this method on several common CNNs, we show that a large portion of the filters can be discarded without obvious accuracy drop, leading to significant reduction of computational burdens. Concretely, we reduce the inference cost of LeNet-5 on MNIST, VGG-16 and ResNet-56 on CIFAR-10 by 95.1%, 79.7% and 60.9%, respectively.