Author name cluster

Zhiming Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

1 author row

AAAI Conference 2026 Conference Paper

KeepKV: Achieving Periodic Lossless KV Cache Compression for Efficient LLM Inference

Yuxuan Tian
Zihan Wang
Yebo Peng
Aomufei Yuan
Zhiming Wang
Bairen Yi
Xin Liu
Yong Cui

Efficient inference of large language models (LLMs) is hindered by an ever-growing key-value (KV) cache, making KV cache compression a critical research direction. Traditional methods selectively evict less important KV cache entries, which leads to information loss and hallucinations. Recently, merging-based strategies have been explored to retain more information by merging KV pairs that would be discarded; however, these existing approaches inevitably introduce inconsistencies in attention distributions before and after merging, causing degraded generation quality. To overcome this challenge, we propose KeepKV, a novel adaptive KV cache merging method designed to preserve performance under strict memory constraints, achieving single-step lossless compression and providing error bounds for multi-step compression. KeepKV introduces the Electoral Votes mechanism that records merging history and adaptively adjusts attention scores. Moreover, it further leverages a novel Zero Inference-Perturbation Merging method, compensating for attention loss resulting from cache merging. Extensive experiments on various benchmarks and LLM architectures demonstrate that KeepKV substantially reduces memory usage while successfully retaining essential context information, achieving over 2 times inference throughput improvement and maintaining superior generation quality even with only 10% KV cache budgets.

PDF Details DOI

AAAI Conference 2025 Conference Paper

TdAttenMix: Top-Down Attention Guided Mixup

Zhiming Wang
Lin Gu
Feng Lu

CutMix is a data augmentation strategy that cuts and pastes image patches to mixup training data. Existing methods pick either random or salient areas which are often inconsistent to labels, thus misguiding the training model. By our knowledge, we integrate human gaze to guide cutmix for the first time. Since human attention is driven by both high-level recognition and low-level clues, we propose a controllable Top-down Attention Guided Module to obtain a general artificial attention which balances top-down and bottom-up attention. The proposed TdATttenMix then picks the patches and adjust the label mixing ratio that focuses on regions relevant to the current label. Experimental results demonstrate that our TdAttenMix outperforms existing state-of-the-art mixup methods across eight different benchmarks. Additionally, we introduce a new metric based on the human gaze and use this metric to investigate the issue of image-label inconsistency.

PDF Details DOI

EAAI Journal 2024 Journal Article

Decision intelligence-driven predictive modelling of air quality index in surface mining

Muhammad Kamran
Izhar Mithal Jiskani
Zhiming Wang
Wei Zhou

Air pollution emanating from both human activities and natural phenomena, is a pervasive and critical issue that affects everyone. In mining, particularly concerning is the detrimental impact of dust contamination, which not only compromises air quality but also jeopardizes sustainability. In order to establish effective measures for preventing and managing mine dust pollution, it is crucial to have a comprehensive understanding of the persistent characteristics associated with air pollution. In response to this challenge, this study presents an innovative decision intelligence-driven predictive modelling of air quality index (AQI) in surface mining. The framework utilizes machine learning approaches, specifically integrating K-means clustering and random forest (RF) algorithms. A comprehensive dataset of 8928 environmental and operational factors impacting air quality was acquired from the Haerwusu Open-pit Coal Mine (HOCM) in China. The AQI was classified into five distinct levels in accordance with the National Ambient Air Quality Standards (NAAQS) established by the United States Environmental Protection Agency (EPA). The application of the K-means clustering algorithm served to reduce the effects of spectral variation in instances with substantial similarities. Subsequently, the RF method was employed to predict varying AQI levels. The findings revealed by the proposed model demonstrate a remarkable accuracy of 97% in predicting the AQI level in surface coal mining. This study provides improved air quality prediction for optimizing mining operation schedules during favorable weather conditions and implementing dust mitigation strategies, thereby enhancing decision-making processes and promoting green and climate-smart mining.

Details DOI

EAAI Journal 2023 Journal Article

Thermal failure of diamond tools indicated by diamond degradation: Damage evaluation and property prediction on small image datasets

Wucheng Sun
Hui Gao
Yuxiang Chen
Zhiming Wang
Longchen Duan
Songcheng Tan
Xiaohong Fang

High temperature induced diamond degradation often leads to the failure of diamond tools. In this work, diamond samples holding different degrees of thermal damage were prepared by heating and sintering. The influence of diamond particle size and processing temperature was investigated through mechanical testing and micromorphology observation, meanwhile, a dataset containing 2870 SEM images showing diamonds with different degrees of degradation was constructed. By modification of VGG16 network, classification models and regression models were developed for thermal damage evaluation and sample property prediction. Training strategies including transfer learning and data augmentation were implemented and verified essential on the small dataset, where drop-out showed no positive effects. Two classification models (3-class and 65-class) were constructed and trained for damage evaluation. Visualized damage feature maps exported from Grad-CAM revealed the influential mechanism of thermal damage on diamonds, which proved the effectiveness of the classification models as well. Under the optimized training strategies, regression models were built for sample property prediction. The models towards toughness index, bending strength loss, relative density and Rockwell hardness were examined. Comparing the output results with real property values in test sets, the first two models matched well, and the latter two showed the opposite. It verified the validity of the regression models for property prediction as they were all established based on diamond damage image datasets. The loss in bending strength loss prediction model was smaller than that of toughness index, indicating bending strength easier to be shorten than impact toughness for diamond/metal composites suffering thermal impacts.

Details DOI