Training machine learning models on sensitive data within cloud environments introduces substantial privacy risks. This paper addresses this challenge by introducing and analyzing a privacy-preserving method for training generative neural networks on obfuscated data. We investigate the security implications of training Conditional Generative Adversarial Networks (CGANs) utilizing an enhanced permutation-based technique. Our method transforms training images via histogram equalization, noise addition, and a pixel permutation function, πimg, while simultaneously obfuscating conditional labels with a function πlabel. Motivated by the need to retain data utility where conventional encryption methods prove insufficient, we demonstrate that this dual approach provides significant operational security for the trained model, M. Specifically, an adversary possessing only the trained model, without knowledge of the functions πimgand πlabel, is computationally prevented from performing meaningful generation. We formalize this notion of security, highlighting its importance for protecting intellectual property and controlling the deployment and usage of trained models.

This paper presents a novel approach to privacy-preserving gesture recognition using a remotely trained neural network. Our method ensures the protection of sensitive user data from potential threats, thereby mitigating concerns about data privacy and security. By utilizing encryption techniques, we enable organizations to train complex machine learning models on large-scale datasets without compromising data integrity. We demonstrate the feasibility of this approach through the implementation of proposed models for gesture classification, which achieved high accuracy in both encrypted and plain modes. Our results show that these models are suitable for embedded devices, making them a viable option for commercial or industrial applications such as smart car navigation systems.

This paper explores privacy-preserving training methods for machine learning models, crucial for protecting sensitive data during cloud-based model training. We propose a novel approach utilizing image encoding and partitioning to train models on remote servers. By partitioning images into encoded patches distributed across servers, each training an independent model, we ensure privacy and resilience against attacks. Our findings demonstrate the feasibility of training private datasets on cloud platforms with minimal accuracy loss, offering a high level of privacy at low cost. A critical aspect of our approach lies in its ability to uphold privacy without any modifications to the training software on cloud servers, unlike methods such as homomorphic encryption, which demand the utilization of specialized software.

Ensuring privacy while outsourcing the training of machine learning (ML) models to cloud-based platforms is a critical concern. Although cryptographic solutions have been proposed, they often result in a substantial reduction in training accuracy and require modifications to the backend architecture. In this paper, we address the challenge of developing privacy-preserving techniques that offer adequate privacy without significantly impacting the accuracy of the ML model or the accuracy of the training process. We demonstrate that training private datasets on existing cloud-based platforms can be achieved with a high level of privacy and at a minimal cost in accuracy.

As part of the responses to the ongoing crypto wars, the notion of Anamorphic Encryption was put forth. The notion allows private communication in spite of a dictator who is engaged in an extreme form of surveillance and or censorship, where it asks for all private keys and knows and may even dictate all messages. The original work pointed out efficient ways to use two known schemes in the anamorphic mode, bypassing the draconian censorship and hiding information from the all-powerful dictator. A question left open was whether these examples are outlier results or whether anamorphic mode is pervasive in existing systems. Here we answer the above question: we develop new techniques, expand the notion, and show that the notion of Anamorphic Cryptography is, in fact, very much prevalent. We first refine the notion of Anamorphic Encryption with respect to the nature of covert communication. Specifically, we distinguish Single-Receiver Encryption for many to one communication, and Multiple-Receiver Encryption for many to many communication within the group of conspiring users. We then show that Anamorphic Encryption can be embedded in the randomness used in the encryption, and we give families of constructions that can be applied to numerous ciphers. In total the families cover classical encryption schemes, some of which in actual use. Among our examples is an anamorphic channel with much higher capacity than the regular channel. In sum, the work shows the very large extent of the potential futility of control and censorship over the use of strong encryption by the dictator (typical for and even stronger than governments engaging in the ongoing crypto-wars): While such limitations obviously hurt utility which encryption typically brings to safety in computing systems, they essentially, are not helping the dictator. While the actual implications of what we show here and what it means in practice require further policy and legal analyses and perspectives, the technical aspects regarding the issues are clearly showing the futility of the war against Cryptography.

We consider the problem of reliable and minimal delay transmission in a wireless sensor network that uses time division in order to schedule its node-to-node communication in time-bounded manner. We propose an algorithm that uses the message acknowledgment method and solves this problem. We show bounds for its expected value of message delivery time. Moreover, our algorithm is based on simple state machine that do not require much computational power, thus could be executed on very weak devices.

Numerical computation on GPU has become easily accessible and offers good computation power for relatively little cost. Recently an application of Newton-Rap son method for analyzing power flow in multi-terminal high-voltage direct current (HVDC) networks was proposed and shown to have good results on five terminal grids. Since this method involves costly matrix operation, especially the inverse, increasing the number of terminals in the grid yields prohibitively large execution times in sequential operation. To address this issue, we adjust the algorithm so that it benefits from parallel computation and test our approach on recent GPU from NVidia. We give experimental results for grids up to few thousand terminals and show that execution time is still acceptable for real applications. We also provide some benchmarks of the GPU computation compared with other platforms.

Efficient cardinality estimation is a common requirement for many wireless sensor network (WSN) applications. The task must be accomplished at extremely low overhead due to severe sensor resource limitation. This poses an interesting challenge for large-scale WSNs. In this paper we present a two-phase probabilistic algorithm based on order statistics and Bernoulli scheme, which effectively estimates the cardinality of WSNs. We thoroughly examine properties of estimators used in each phase as well as the precision of the whole procedure. The algorithm discussed in this paper is a modification of a recently published idea - the modification enables us to obtain a provable precision.

We consider an ad-hoc network consisting of devices that try to gain access for transmission through a shared radio communication channel. We consider two randomized leader election protocols the first one is due to Nakanoand Olariu (2000); the second one is due to Cai, Lu and Wang (2003) and propose combinations which give us an improvement of both of them. We show that with discrete starting points of transmission, between which a station may choose in a non-uniform way, leads to a simple algorithm that substantially outperforms the previous techniques of resolving channel access problems. We provide methods to optimize values of parameters used.