Jie Zhang

Lei Zhang, Jie Zhang, Bowen Lei et al. · microsoft-research

Dataset Distillation (DD), a newly emerging field, aims at generating much smaller but efficient synthetic training datasets from large ones. Existing DD methods based on gradient matching achieve leading performance; however, they are extremely computationally intensive as they require continuously optimizing a dataset among thousands of randomly initialized models. In this paper, we assume that training the synthetic data with diverse models leads to better generalization performance. Thus we propose two model augmentation techniques, i.e. using early-stage models and parameter perturbation to learn an informative synthetic set with significantly reduced training cost. Extensive experiments demonstrate that our method achieves up to 20x speedup and comparable performance on par with state-of-the-art methods.

Zhongqi Wang, Jie Zhang, Zhilong Ji et al.

With the development of deep generative models, recent years have seen great success of Chinese landscape painting generation. However, few works focus on controllable Chinese landscape painting generation due to the lack of data and limited modeling capabilities. In this work, we propose a controllable Chinese landscape painting generation method named CCLAP, which can generate painting with specific content and style based on Latent Diffusion Model. Specifically, it consists of two cascaded modules, i.e., content generator and style aggregator. The content generator module guarantees the content of generated paintings specific to the input text. While the style aggregator module is to generate paintings of a style corresponding to a reference image. Moreover, a new dataset of Chinese landscape paintings named CLAP is collected for comprehensive evaluation. Both the qualitative and quantitative results demonstrate that our method achieves state-of-the-art performance, especially in artfully-composed and artistic conception. Codes are available at https://github.com/Robin-WZQ/CCLAP.

Jianhao Yuan, Jie Zhang, Shuyang Sun et al.

Synthetic training data has gained prominence in numerous learning tasks and scenarios, offering advantages such as dataset augmentation, generalization evaluation, and privacy preservation. Despite these benefits, the efficiency of synthetic data generated by current methodologies remains inferior when training advanced deep models exclusively, limiting its practical utility. To address this challenge, we analyze the principles underlying training data synthesis for supervised learning and elucidate a principled theoretical framework from the distribution-matching perspective that explicates the mechanisms governing synthesis efficacy. Through extensive experiments, we demonstrate the effectiveness of our synthetic data across diverse image classification tasks, both as a replacement for and augmentation to real datasets, while also benefits such as out-of-distribution generalization, privacy preservation, and scalability. Specifically, we achieve 70.9% top1 classification accuracy on ImageNet1K when training solely with synthetic data equivalent to 1 X the original real data size, which increases to 76.0% when scaling up to 10 X synthetic data.

Leijie Wu, Song Guo, Yaohong Ding et al.

Self-attention mechanisms, especially multi-head self-attention (MSA), have achieved great success in many fields such as computer vision and natural language processing. However, many existing vision transformer (ViT) works simply inherent transformer designs from NLP to adapt vision tasks, while ignoring the fundamental difference between ``how MSA works in image and language settings''. Language naturally contains highly semantic structures that are directly interpretable by humans. Its basic unit (word) is discrete without redundant information, which readily supports interpretable studies on MSA mechanisms of language transformer. In contrast, visual data exhibits a fundamentally different structure: Its basic unit (pixel) is a natural low-level representation with significant redundancies in the neighbourhood, which poses obvious challenges to the interpretability of MSA mechanism in ViT. In this paper, we introduce a typical image processing technique, i.e., scale-invariant feature transforms (SIFTs), which maps low-level representations into mid-level spaces, and annotates extensive discrete keypoints with semantically rich information. Next, we construct a weighted patch interrelation analysis based on SIFT keypoints to capture the attention patterns hidden in patches with different semantic concentrations Interestingly, we find this quantitative analysis is not only an effective complement to the interpretability of MSA mechanisms in ViT, but can also be applied to 1) spurious correlation discovery and ``prompting'' during model inference, 2) and guided model pre-training acceleration. Experimental results on both applications show significant advantages over baselines, demonstrating the efficacy of our method.

Yunwen Xia, Hui Fang, Jie Zhang et al.

Conversational recommender system (CRS), which combines the techniques of dialogue system and recommender system, has obtained increasing interest recently. In contrast to traditional recommender system, it learns the user preference better through interactions (i.e. conversations), and then further boosts the recommendation performance. However, existing studies on CRS ignore to address the relationship among attributes, users, and items effectively, which might lead to inappropriate questions and inaccurate recommendations. In this view, we propose a knowledge graph based conversational recommender system (referred as KG-CRS). Specifically, we first integrate the user-item graph and item-attribute graph into a dynamic graph, i.e., dynamically changing during the dialogue process by removing negative items or attributes. We then learn informative embedding of users, items, and attributes by also considering propagation through neighbors on the graph. Extensive experiments on three real datasets validate the superiority of our method over the state-of-the-art approaches in terms of both the recommendation and conversation tasks.

Yingpeng Du, Ziyan Wang, Zhu Sun et al.

In recent years, efforts have been made to use text information for better user profiling and item characterization in recommendations. However, text information can sometimes be of low quality, hindering its effectiveness for real-world applications. With knowledge and reasoning capabilities capsuled in Large Language Models (LLMs), utilizing LLMs emerges as a promising way for description improvement. However, existing ways of prompting LLMs with raw texts ignore structured knowledge of user-item interactions, which may lead to hallucination problems like inconsistent description generation. To this end, we propose a Graph-aware Convolutional LLM method to elicit LLMs to capture high-order relations in the user-item graph. To adapt text-based LLMs with structured graphs, We use the LLM as an aggregator in graph processing, allowing it to understand graph-based information step by step. Specifically, the LLM is required for description enhancement by exploring multi-hop neighbors layer by layer, thereby propagating information progressively in the graph. To enable LLMs to capture large-scale graph information, we break down the description task into smaller parts, which drastically reduces the context length of the token input with each step. Extensive experiments on three real-world datasets show that our method consistently outperforms state-of-the-art methods.

Yingpeng Du, Zhu Sun, Ziyan Wang et al.

Large language models (LLMs) provide a promising way for accurate session-based recommendation (SBR), but they demand substantial computational time and memory. Knowledge distillation (KD)-based methods can alleviate these issues by transferring the knowledge to a small student, which trains a student based on the predictions of a cumbersome teacher. However, these methods encounter difficulties for \textit{LLM-based KD in SBR}. 1) It is expensive to make LLMs predict for all instances in KD. 2) LLMs may make ineffective predictions for some instances in KD, e.g., incorrect predictions for hard instances or similar predictions as existing recommenders for easy instances. In this paper, we propose an active LLM-based KD method in SBR, contributing to sustainable AI. To efficiently distill knowledge from LLMs with limited cost, we propose to extract a small proportion of instances predicted by LLMs. Meanwhile, for a more effective distillation, we propose an active learning strategy to extract instances that are as effective as possible for KD from a theoretical view. Specifically, we first formulate gains based on potential effects (e.g., effective, similar, and incorrect predictions by LLMs) and difficulties (e.g., easy or hard to fit) of instances for KD. Then, we propose to maximize the minimal gains of distillation to find the optimal selection policy for active learning, which can largely avoid extracting ineffective instances in KD. Experiments on real-world datasets show that our method significantly outperforms state-of-the-art methods for SBR.

Congqing He, Jie Zhang, Xiangyu Zhu et al.

Consumer Event-Cause Extraction, the task aimed at extracting the potential causes behind certain events in the text, has gained much attention in recent years due to its wide applications. The ICDM 2020 conference sets up an evaluation competition that aims to extract events and the causes of the extracted events with a specified subject (a brand or product). In this task, we mainly focus on how to construct an end-to-end model, and extract multiple event types and event-causes simultaneously. To this end, we introduce a fresh perspective to revisit the relational event-cause extraction task and propose a novel sequence tagging framework, instead of extracting event types and events-causes separately. Experiments show our framework outperforms baseline methods even when its encoder module uses an initialized pre-trained BERT encoder, showing the power of the new tagging framework. In this competition, our team achieved 1st place in the first stage leaderboard, and 3rd place in the final stage leaderboard.

Junyang Lin, An Yang, Jinze Bai et al.

Recent expeditious developments in deep learning algorithms, distributed training, and even hardware design for large models have enabled training extreme-scale models, say GPT-3 and Switch Transformer possessing hundreds of billions or even trillions of parameters. However, under limited resources, extreme-scale model training that requires enormous amounts of computes and memory footprint suffers from frustratingly low efficiency in model convergence. In this paper, we propose a simple training strategy called "Pseudo-to-Real" for high-memory-footprint-required large models. Pseudo-to-Real is compatible with large models with architecture of sequential layers. We demonstrate a practice of pretraining unprecedented 10-trillion-parameter model, an order of magnitude larger than the state-of-the-art, on solely 512 GPUs within 10 days. Besides demonstrating the application of Pseudo-to-Real, we also provide a technique, Granular CPU offloading, to manage CPU memory for training large model and maintain high GPU utilities. Fast training of extreme-scale models on a decent amount of resources can bring much smaller carbon footprint and contribute to greener AI.

Junyang Lin, Rui Men, An Yang et al.

In this work, we construct the largest dataset for multimodal pretraining in Chinese, which consists of over 1.9TB images and 292GB texts that cover a wide range of domains. We propose a cross-modal pretraining method called M6, referring to Multi-Modality to Multi-Modality Multitask Mega-transformer, for unified pretraining on the data of single modality and multiple modalities. We scale the model size up to 10 billion and 100 billion parameters, and build the largest pretrained model in Chinese. We apply the model to a series of downstream applications, and demonstrate its outstanding performance in comparison with strong baselines. Furthermore, we specifically design a downstream task of text-guided image generation, and show that the finetuned M6 can create high-quality images with high resolution and abundant details.

Yanshi Wang, Jie Zhang, Qing Da et al.

Estimating post-click conversion rate (CVR) accurately is crucial in E-commerce. However, CVR prediction usually suffers from three major challenges in practice: i) data sparsity: compared with impressions, conversion samples are often extremely scarce; ii) sample selection bias: conventional CVR models are trained with clicked impressions while making inference on the entire space of all impressions; iii) delayed feedback: many conversions can only be observed after a relatively long and random delay since clicks happened, resulting in many false negative labels during training. Previous studies mainly focus on one or two issues while ignoring the others. In this paper, we propose a novel neural network framework ESDF to tackle the above three challenges simultaneously. Unlike existing methods, ESDF models the CVR prediction from a perspective of entire space, and combines the advantage of user sequential behavior pattern and the time delay factor. Specifically, ESDF utilizes sequential behavior of user actions on the entire space with all impressions to alleviate the sample selection bias problem. By sharing the embedding parameters between CTR and CVR networks, data sparsity problem is greatly relieved. Different from conventional delayed feedback methods, ESDF does not make any special assumption about the delay distribution. We discretize the delay time by day slot and model the probability based on survival analysis with deep neural network, which is more practical and suitable for industrial situations. Extensive experiments are conducted to evaluate the effectiveness of our method. To the best of our knowledge, ESDF is the first attempt to unitedly solve the above three challenges in CVR prediction area.

Juyong Jiang, Jie Zhang, Kai Zhang

Transformers have shown great success in learning representations for language modelling. However, an open challenge still remains on how to systematically aggregate semantic information (word embedding) with positional (or temporal) information (word orders). In this work, we propose a new architecture to aggregate the two sources of information using cascaded semantic and positional self-attention network (CSPAN) in the context of document classification. The CSPAN uses a semantic self-attention layer cascaded with Bi-LSTM to process the semantic and positional information in a sequential manner, and then adaptively combine them together through a residue connection. Compared with commonly used positional encoding schemes, CSPAN can exploit the interaction between semantics and word positions in a more interpretable and adaptive manner, and the classification performance can be notably improved while simultaneously preserving a compact model size and high convergence rate. We evaluate the CSPAN model on several benchmark data sets for document classification with careful ablation studies, and demonstrate the encouraging results compared with state of the art.

Jie Zhang, Xiaolong Wang, Dawei Li et al.

State-of-the-art deep model compression methods exploit the low-rank approximation and sparsity pruning to remove redundant parameters from a learned hidden layer. However, they process each hidden layer individually while neglecting the common components across layers, and thus are not able to fully exploit the potential redundancy space for compression. To solve the above problem and enable further compression of a model, removing the cross-layer redundancy and mining the layer-wise inheritance knowledge is necessary. In this paper, we introduce a holistic model compression framework, namely MIning Cross-layer Inherent similarity Knowledge (MICIK), to fully excavate the potential redundancy space. The proposed MICIK framework simultaneously, (1) learns the common and unique weight components across deep neural network layers to increase compression rate; (2) preserves the inherent similarity knowledge of nearby layers and distant layers to minimize the accuracy loss and (3) can be complementary to other existing compression techniques such as knowledge distillation. Extensive experiments on large-scale convolutional neural networks demonstrate that MICIK is superior over state-of-the-art model compression approaches with 16X parameter reduction on VGG-16 and 6X on GoogLeNet, all without accuracy loss.