Sharing, Licensing, Buying, Selling and Operationalizing ML Models: A Deep Learning based Co-operative and Co-ordinated Security usecase

• Published in: Australasian Telecommunication Networks and Applications Conference [proceedings] pp. 118 - 123
• Main Authors: Panchal, Deven, Musgrove, Dan, Baran, Isilay, Lu, David
• Format: Conference Proceeding
• Language: English
• Published: IEEE 29.11.2023
• Subjects: AI4EU; CCPA; Companies; Computational modeling; Cooperative ML; Cooperative Security; Data Location; Data models; Data transfer; DDoS; Deep Learning; Denial-of-service attack; Federated Learning; GDPR; Machine Learning; ML model federation; ML model licensing; ML Model monetization; ML Model Sharing; Open Source; Training; Writing
• ISSN: 2474-154X
• DOI: 10.1109/ITNAC59571.2023.10368564

Many problems that utilize Machine Learning in fact require multiple cooperative efforts to train models on different data on different targets so as to then be able to utilize all these results or the models themselves to do certain tasks. Problems in Remote sensing, Cybersecurity, Network analytics lend themselves very well to this co-operative paradigm where models built elsewhere on other data can be of direct and immediate use or at least be retrained and used at a different site. These models could in fact also come from some other organization or company (model supplier) who might charge (monetize) the usage of these models to another organization or company (model consumer) who may want to deploy them for their own use. Model sharing becomes all the more important in scenarios in which the data is subject to data sovereignty and data location requirements and, requirements over data transfer like GDPR, CCPA, etc. This sharing, monetizing, licensing, writing terms of sale of the models can be done in a very easy and streamlined fashion using the Acumos Federation and Acumos Licensing components that we have built in Acumos. To demonstrate their usage, we build an ML model to classify Distributed Denial of Service (DDoS) attacks and then set up N-site wide federation across 'N' Acumos instances (N=3 in the paper for demo purposes, which could represent 3 companies or 3 different physical sites) to share and use the models built elsewhere to classify DDoS attack types. This paper shows how we can set up a coordinated and cooperative defense against a cooperative attack like DDoS, and more generally solve a variety of problems across domains using ML in a cooperative fashion with Acumos.