Call for Papers - Special Issue on Secured Machine Learning Framework Using Multi-Path Network Aggregation
Scalable machine learning systems analysis of information acquired from decentralized consumer equipment has grown rapidly in popularity over the years. Unfortunately, most effective machine learning techniques on such a community scale may not yet link inherent value, governance, and confidentiality with users' information privacy. There is an underlying dispute among two methodological approaches: conventional centralized and decentralized training. Whereas centralized training infringes the fundamental liberties of participants, viable training strategies are often insufficiently adequate. Individuals with malevolent purposes might damage the training model by purposely giving it inaccurate information, data poisoning. When applied to datasets including private information, rules such as General Data Protection Regulation necessitate the employment of privacy-preserving approaches.
Furthermore, if seen like multi-party computation (MPC), aggregating in machine learning somehow does not maintain input confidentiality since the device sees the prototype-based. To maintain input confidentiality, every partner must understand nothing except the computation result, and in this scenario seems to be the aggregation model improvements. In order to address the issue mentioned earlier, reliable aggregation procedures are necessary to ensure input confidentiality. Certain safe aggregation principles have lately gained widespread in industrial usage, where input privacy assurances may provide efficiency and precision advantages over differential privacy. Due to the ever-increasing need for efficient machine learning (ML) modeling techniques, deep neural networks' dataset and prototype dimensions have increased steadily. Consequently, training current DNN frameworks exceeds the capability of a single system, and building big frameworks now requires hundreds of accelerators.
Additionally, even as ubiquity and diversity of neural network prototypes increases, including first party and third party trainee work demands, they are progressively moving to distributed clusters extending hundreds of rows and performing several different machine learning tasks. These activities cover basic machine learning tasks like K-means clustering and more modern ones like deep learning and reinforcement learning. Besides an effective method for sharing system resources at a wide scale, the operational feasibility of Multi-Path Network aggregation in much more plausible situations appears doubtful. The aggregation results in a diverse group of streams of various sizes from those couple KBs up to tens of GBs, yet only a subset may need aggregation. Furthermore, difficulties arise from sophisticated processing correlations, constant intra-node information transfer between the GPU and CPU, and extensive inter-node information interaction. In combination with an ultra-light congestion management protocol, the load-balancing technique disseminates aggregation information over many aggregation nodes, reducing bandwidth problem areas and improving the network's performance.
This special issue's objective is to publish original research papers on the Machine Learning Framework for secured multi-path network aggregations in order to reduce bandwidth congestion and enhance network performance. This comprises research targeted at resolving issues brought about by complex processing correlations, continuous information flow between the GPU and CPU, and intensive information interaction between nodes.
Topics include, but are not limited to:
- Analysis Communication overhead in Multi-Path Network Aggregation for Secured Machine Learning Framework
- Advanced fault tolerance mechanism for securing Machine Learning Framework
- Investigation of Compatibility issues to guarantee user-level privacy in securing Machine Learning Framework
- Mitigation privacy issue in Machine learning as a service framework
- Enhanced cryptographic methods for Multi-Path Network Aggregation in Secured Machine Learning Framework
- Convergence of blockchain with Machine Learning Framework for enhancing security constraints
- Privacy enhanced computational intelligence in Multi-Path Network Aggregation for Securing Machine Learning Framework
- Advanced Clustering techniques in Multi-Path Network Aggregation for Securing Machine Learning Framework
- Novel load-balancing technique in Multi-Path Network Aggregation for Secured Machine Learning Framework
- Extensive inter-node information interaction in Multi-Path Network Aggregation for Secured Machine Learning Framework
Proposed Special Issue Timeline:
Manuscript submission deadline: January 31, 2023
Authors notification: April 25, 2023
Revised papers due: June 15, 2023
Final notification: September 20, 2023
J. Alfred Daniel, Karpagam Academy of Higher Education, Coimbatore, India
Awais Ahmad, Università Degli Studi di Milano, Milan, Italy
Boris Tomaš, University of Zagreb, Varaždin, Croatia
C. Chandru Vignesh, Vel Tech Rangarajan Dr.Sagunthala R&D Institute of Science and Technology, Chennai, Tamil Nadu, India