Xinyao Liu

Xinyao Liu, Shengdong Du, Tianrui Li et al.

With the advent of the big data era, the data quality problem is becoming more critical. Among many factors, data with missing values is one primary issue, and thus developing effective imputation models is a key topic in the research community. Recently, a major research direction is to employ neural network models such as self-organizing mappings or automatic encoders for filling missing values. However, these classical methods can hardly discover interrelated features and common features simultaneously among data attributes. Especially, it is a very typical problem for classical autoencoders that they often learn invalid constant mappings, which dramatically hurts the filling performance. To solve the above-mentioned problems, we propose a missing-value-filling model based on a feature-fusion-enhanced autoencoder. We first incorporate into an autoencoder a hidden layer that consists of de-tracking neurons and radial basis function neurons, which can enhance the ability of learning interrelated features and common features. Besides, we develop a missing value filling strategy based on dynamic clustering that is incorporated into an iterative optimization process. This design can enhance the multi-dimensional feature fusion ability and thus improves the dynamic collaborative missing-value-filling performance. The effectiveness of the proposed model is validated by extensive experiments compared to a variety of baseline methods on thirteen data sets.

Guang Liang, Xinyao Liu, Jianxin Wu

Vision Transformers (ViTs) are essential in computer vision but are computationally intensive, too. Model quantization, particularly to low bit-widths like 4-bit, aims to alleviate this difficulty, yet existing Post-Training Quantization (PTQ) and Quantization-Aware Training (QAT) methods exhibit significant limitations. PTQ often incurs substantial accuracy drop, while QAT achieves high accuracy but suffers from prohibitive computational costs, limited generalization to downstream tasks, training instability, and lacking of open-source codebase. To address these challenges, this paper introduces General, Practical, and Lightning Quantization (GPLQ), a novel framework designed for efficient and effective ViT quantization. GPLQ is founded on two key empirical insights: the paramount importance of activation quantization and the necessity of preserving the model's original optimization ``basin'' to maintain generalization. Consequently, GPLQ employs a sequential ``activation-first, weights-later'' strategy. Stage 1 keeps weights in FP32 while quantizing activations with a feature mimicking loss in only 1 epoch to keep it stay in the same ``basin'', thereby preserving generalization. Stage 2 quantizes weights using a PTQ method. As a result, GPLQ is 100x faster than existing QAT methods, lowers memory footprint to levels even below FP32 training, and achieves 4-bit model performance that is highly competitive with FP32 models in terms of both accuracy on ImageNet and generalization to diverse downstream tasks, including fine-grained visual classification and object detection. We will release an easy-to-use open-source toolkit supporting multiple vision tasks.

Guang Liang, Jie Shao, Ningyuan Tang et al.

Native FP8 support in modern hardware is essential for training large Transformers, but is severely hindered by extreme activation outliers. Existing solutions either rely on complex mixed-precision engineering or invasive architectural modifications. This paper fundamentally challenges the conventional wisdom that outliers are data-driven. We demonstrate that extreme outliers are a data-independent, mechanically-produced artifact of training, originating from specific structural properties of the weight matrices (i.e., colinearity). Based on this insight, we propose TWEO (Transformers Without Extreme Outliers), a novel, non-invasive loss function. TWEO effectively prevents extreme outliers via a very simple loss term, which reduces outliers from 10000+ to less than 20. TWEO then enables full-model FP8 pre-training with neither engineering tricks nor architectural changes for both LLM and ViT. When standard FP8 training catastrophically collapses, TWEO achieves performance comparable to the BF16 baseline while delivering a 36% increase in training throughput. Also, TWEO enables a new quantization paradigm. Hardware-friendly W8A8 per-tensor static quantization of LLMs, previously considered completely unusable due to outliers, achieves SOTA performance for the first time on TWEO-trained models.