Privacy preservation for Neural Network Training
Journal Title: International Journal of Computer Science & Engineering Technology - Year 2014, Vol 5, Issue 6
Abstract
Data mining enables us to extract interesting patterns from huge volumes of data. This can be a security/privacy issue, if we make sensitive information available to unintended users as a result of mining. Preserving the privacy of sensitive data and individuals' information is a major challenge in many applications. Privacy concerns may prevent the parties from directly sharing the data and some types of information about the data. Back propagation (BP) is one of the algorithms used for training neural networks. This algorithm is suitable for both single-layer and multi-layer models. Another neural network training algorithm values for weights. Training the network is achieved by iteratively adjusting the weights based on a set of input parameters. During this process a training sample is presented to the network and is forwarded to determine the resulting output. The difference between the obtained output and expected output in each output unit represents error. Error is back propagated is the Extreme Learning Machine (ELM) algorithm. This algorithm works only for single-layer models. This restriction makes ELM more efficient than BP by reducing the communication between multiple parties during the learning phase. Most of the neural network learning systems is designed for single-layer and multi-layer models which can be applied to continuous data and differentiable activation functions. The present work proposes a protocol for neural network training systems that guarantee data confidentiality when data is partitioned among several parties. The proposed protocol preserves the privacy of both the input data and the constructed learning model.
Authors and Affiliations
Ms. Ch. Aparna , Ms. K. Venkata Ramana
Keywords
Related Articles
- EP ID EP116128
- DOI -
- Views 96
- Downloads 0
Impact of Protocol Switching Strategy in Delay Tolerant Networks
Delay-Tolerant Networks (DTNs) are sparse dynamic wireless networks in which a complete end-to-end path from the source to the destination does not exist. In this context, conventional mobile adhoc routing schemes would...
Evaluating Recommender Strategies
Recommender systems are a subclass of information filtering systems that seek to generate meaningful recommendations to users for products or items that might interest them. In recent times, it has become common to colle...
Named Entity Recognition in Punjabi Using Hidden Markov Model
Named Entity Recognition (NER) is a task to discover the Named Entities (NEs) in a document and then categorize these NEs into diverse Named Entity classes such as Name of Person, Location, River, Organization etc. Since...
Secure Data Sharing using Decoy Technology
Cloud computing is a virtualized compute power and storage delivered via platform-agnostic infrastructures of abstracted hardware and software accessed over the Internet. Data sharing is an important functionality in clo...
A NOVEL APPROACH FOR COMMUNITY DISCOVERY IN DYNAMIC NETWORKS
Recently, discovering aggressive communities has become an increasingly critical task. Many conclusion have been expected, most of which only use correlation structure. However, rich information is cipher in the content...