Wed 6 Dec 2023 15:15 - 15:30 at Golden Gate C2 - Machine Learning III Chair(s): Rangeet Pan

The use of modern Natural Language Processing (NLP) techniques has shown to be beneficial for software engineering tasks, such as vulnerability detection and type inference. However, training deep NLP models requires significant computational resources. This paper explores techniques that aim at achieving the best usage of resources and available information in these models.

We propose a generic approach, EarlyBIRD, to build composite representations of code from the early layers of a pre-trained transformer model. We empirically investigate the viability of this approach on the CodeBERT model by comparing the performance of 12 strategies for creating composite representations with the standard practice of only using the last encoder layer.

Our evaluation on four datasets shows that several early layer combinations yield better performance on defect detection, and some combinations improve multi-class classification. More specifically, we obtain a +2 average improvement of detection accuracy on Devign with only 3 out of 12 layers of CodeBERT and a 3.3x speed-up of fine-tuning. These findings show that early layers can be used to obtain better results using the same resources, as well as to reduce resource usage during fine-tuning and inference.

Wed 6 Dec

Displayed time zone: Pacific Time (US & Canada) change

14:00 - 15:30
Machine Learning IIIDemonstrations / Industry Papers / Research Papers at Golden Gate C2
Chair(s): Rangeet Pan IBM Research
14:00
15m
Talk
Benchmarking Robustness of AI-enabled Multi-sensor Fusion Systems: Challenges and Opportunities
Research Papers
XinyuGao , Zhijie Wang University of Alberta, Yang Feng Nanjing University, Lei Ma The University of Tokyo / University of Alberta, Zhenyu Chen Nanjing University, Baowen Xu Nanjing University
Media Attached
14:15
7m
Talk
A Language Model of Java Methods with Train/Test Deduplication
Demonstrations
Chia-Yi Su University of Notre Dame, Aakash Bansal University of Notre Dame, Vijayanta Jain University of Maine, Sepideh Ghanavati University of Maine , Collin McMillan University of Notre Dame
Media Attached
14:23
7m
Talk
DENT - A Tool for Tagging Stack Overflow Posts With Deep Learning Energy Patterns
Demonstrations
Shriram Shanbhag Indian Institute of Technology Tirupati, Sridhar Chimalakonda Indian Institute of Technology Tirupati, Vibhu Saujanya Sharma Accenture Labs, India, Vikrant Kaulgud Accenture Labs, India
Media Attached
14:30
15m
Talk
Automated Testing and Improvement of Named Entity Recognition Systems
Research Papers
BoXi Yu The Chinese University of Hong Kong, Shenzhen, Yiyan Hu The Chinese University of Hong Kong, Shenzhen, Qiuyang Mang The Chinese University of Hong Kong, Shenzhen, Wenhan Hu The Chinese University of Hong Kong, Shenzhen, Pinjia He The Chinese University of Hong Kong, Shenzhen
Pre-print Media Attached
14:45
15m
Talk
KDDT: Knowledge Distillation-Empowered Digital Twin for Anomaly Detection
Industry Papers
Xu Qinghua Simula Research Laboratory; University of Oslo, Shaukat Ali Simula Research Laboratory and Oslo Metropolitan University, Tao Yue Beihang University, Zaimovic Nedim Alstom Rail, Inderjeet Singh Alstom
DOI Media Attached
15:00
15m
Talk
Deep Learning Based Feature Envy Detection Boosted by Real-World Examples
Research Papers
Bo Liu Beijing Institute of Technology, Hui Liu Beijing Institute of Technology, Guangjie Li National Innovation Institute of Defense Technology, Nan Niu University of Cincinnati, Zimao Xu Beijing Institute of Technology, Yifan Wang Huawei Cloud, Yunni Xia Chongqing University, Yuxia Zhang Beijing Institute of Technology, Yanjie Jiang Peking University
DOI Pre-print Media Attached
15:15
15m
Talk
[Remote] The EarlyBIRD Catches the Bug: On Exploiting Early Layers of Encoder Models for More Efficient Code Classification
Research Papers
Anastasiia Grishina Simula Research Laboratory, Max Hort Simula Research Laboratory, Leon Moonen Simula Research Laboratory and BI Norwegian Business School
Pre-print Media Attached