Grant: Alliance Grants
University of Alberta
$56,000.00
Malware detection, classification and analysis through continual learning of deep neural networks and large language models
FY 2024-25
Summary
There is a growing demand in strengthening cybersecurity industry with latest AI technologies. This project proposes the development of a full framework of AI-driven anti-malware solution, by focusing on five problems impeding the deployment of ML-based malware detection/analysis in the real world: 1) periodic model updates to address constantly changing and emerging threats, where model training on new data must not compromise detection performance on previously seen malware types; 2) enable distributed and federated model training based on proprietary datasets owned by different clients; 3) malware classification model personalization per client environment, customized to client resources, computing hardware and individual task heterogeneity; 4) generate human language-based explanations and analysis for malicious code using large language models (LLMs); 5) the training of lightweight LLM for client/edge deployment through customization to device resources. Objectives 1, 2 will implement a novel Continual Learning framework using Mixture of Experts (MoE) architectures to train malware classification models, while enabling federated model training and aggregation to leverage proprietary client datasets. Objective 3 will investigate the use of Neural Architecture Search, knowledge distillation, MoE pruning, and other methods to customize client models. Objectives 4, 5 will investigate malware code analysis generation using Large Language Models (e.g., GPT4) via prompt engineering and the distillation of lightweight LLMs that are customized to client resource constraints by mimicking large model behaviors.
The proposed framework will be deployed and tested in collaboration with Wedge Networks, a Calgary-based Canadian cybersecurity company. The anticipated outcome is a novel fully automated AI-driven anti-malware detection, classification, and analysis service, which augments traditional anti-malware practice, thereby providing the company and Alberta an advantage over competitors globally in the transition of cybersecurity technology.
Program
Alliance Grants
Recipient
Location
Alberta, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
598406-2024
Name nom
Niu, Di D
Department département
Electrical and Computer Engineering
Organization id
9
Institution établissement
University of Alberta
Province en
Alberta
Province fr
Alberta
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
56000
Program id
ALLRP
Program name en
Alliance Grants
Program name fr
Subventions Alliance
Group en
RESEARCH PARTNERSHIPS
Group fr
PARTENARIATS DE RECHERCHE
Committee code
96
Committee name en
RPP Internal Decision Cttee
Committee name fr
Cté de décision interne - PPR
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2800
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Artificial intelligence (computer vision, use2603)
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Malware detection, classification and analysis through continual learning of deep neural networks and large language models
Keywords
continual learning; cybersecurity; federated learning; knowledge distillation; large language model; malware analysis; malware classification; mixture-of-experts; model compression; neural architecture search
Application summary
There is a growing demand in strengthening cybersecurity industry with latest AI technologies. This project proposes the development of a full framework of AI-driven anti-malware solution, by focusing on five problems impeding the deployment of ML-based malware detection/analysis in the real world: 1) periodic model updates to address constantly changing and emerging threats, where model training on new data must not compromise detection performance on previously seen malware types; 2) enable distributed and federated model training based on proprietary datasets owned by different clients; 3) malware classification model personalization per client environment, customized to client resources, computing hardware and individual task heterogeneity; 4) generate human language-based explanations and analysis for malicious code using large language models (LLMs); 5) the training of lightweight LLM for client/edge deployment through customization to device resources. Objectives 1, 2 will implement a novel Continual Learning framework using Mixture of Experts (MoE) architectures to train malware classification models, while enabling federated model training and aggregation to leverage proprietary client datasets. Objective 3 will investigate the use of Neural Architecture Search, knowledge distillation, MoE pruning, and other methods to customize client models. Objectives 4, 5 will investigate malware code analysis generation using Large Language Models (e.g., GPT4) via prompt engineering and the distillation of lightweight LLMs that are customized to client resource constraints by mimicking large model behaviors.
The proposed framework will be deployed and tested in collaboration with Wedge Networks, a Calgary-based Canadian cybersecurity company. The anticipated outcome is a novel fully automated AI-driven anti-malware detection, classification, and analysis service, which augments traditional anti-malware practice, thereby providing the company and Alberta an advantage over competitors globally in the transition of cybersecurity technology.