CCCristina Comaniciu

Cristina Comaniciu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Cristina Comaniciu has a total of 23 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Pseudoinvertible Neural Networks

    by Elijah D. Bolluyt, Cristina Comaniciu

    This work introduces novel architecture components and training procedures to create augmented neural networks with the ability to process data bidirectionally via an end-to-end approximate inverse. We develop Pseudo-Invertible Neural Network (PsI-NN) layers which function as drop-in replacements for corresponding convolutional and fully connected layers; by using these, existing architectures gain a pseudoinverse function which, with training, approximately reverses the forward function. For cases where learning both a task and its inverse are necessary or desirable, we show that PsI-NN enabled models match or exceed the quality of results generated by systems that use two separate models while drastically reducing system parameter count. We demonstrate this on two tasks: unpaired image translation and semi-supervised classification. In both, PsI-NN greatly reduces parameter count without any loss of output quality; in semi supervised image classification, PsI-NN improves classification performance beyond the baseline autoencoder method. Our approach creates pseudo-invertible versions of existing architectures, circumventing the stringent constraints required to create a true inverse while allowing a single neural network to learn a task in two directions simultaneously.


  • TONet: A Fast and Efficient Method for Traffic Obfuscation Using Adversarial Machine Learning

    by Fan Yang, Bingyang Wen, Cristina Comaniciu, K. P. Subbalakshmi, R. Chandramouli

    In this paper, we address the problem of privacy leakage in communications based on analysis of traffic patterns. We propose an efficient method of traffic obfuscation based on neural networks, that generates traffic distortions with minimal overhead and computational cost. Our experimental results show that the proposed method is orders of magnitude faster in implementation and has a higher obfuscation success rate with less perturbation on the traffic samples, compared to previously proposed adversarial machine learning-based traffic obfuscation methods.


  • Open Access

    Collapse Resistant Deep Convolutional GAN for Multi-Object Image Generation

    by Elijah D. Bolluyt, Cristina Comaniciu

    This work introduces a novel system for the generation of images that contain multiple classes of objects. Recent work in Generative Adversarial Networks have produced high quality images, but many focus on generating images of a single object or set of objects. Our system addresses the task of image generation conditioned on a list of desired classes to be included in a single image. This enables our system to generate images with any given combination of objects, all composed into a visually realistic natural image. The system learns the interrelationships of all classes represented in a dataset, and can generate diverse samples including a set of these classes. It displays the ability to arrange these objects together, accounting for occlusions and inter-object spatial relations that characterize complex natural images. To accomplish this, we introduce a novel architecture based on Conditional Deep Convolutional GANs that is stabilized against collapse relative to both mode and condition. The system learns to rectify mode collapse during training, self-correcting to avoid suboptimal generation modes.

  • Dynamic Influence on Replicator Evolution for the Propagation of Competing Technologies

    by Elijah D. Bolluyt, Cristina Comaniciu

    This work introduces a novel modified Replicator Dynamics model, which includes external influences on the population. This framework models a realistic market into which companies, the external dynamic influences, invest resources in order to bolster their product’s standing and increase their market share. The dynamic influences change in each time step of the game, and directly modify the payoff matrix of the population’s interactions. The model can learn from real data how each influence affects the market, and can be used to simulate and predict the outcome of a real system. We specifically analyze how a new technology can compete and attempt to unseat an entrenched technology as the market leader. We establish a relationship between the external influences and the population payoff matrix and show how the system can be implemented to predict outcomes in a real market by simulating the rise of the Android mobile operating system over its primary competition, the iPhone, from 2009-2017.


  • On Energy-Secrecy Trade-Offs for Gaussian Wiretap Channels

    by Cristina Comaniciu, H. Vincent Poor

    This paper investigates physical-layer security-energy trade-offs from an information theoretic perspective. The metric of partial secrecy is introduced as the amount of distortion that can be guaranteed at the eavesdropper, and three different channel models are proposed to capture various transmission strategies to achieve partial secrecy. It is shown that a range of secrecy-energy operating points can be achieved with an appropriate allocation of power resources between private and nonprivate data substreams.


  • Toward Increasing Packet Diversity for Relaying LT Fountain Codes in Wireless Sensor Networks

    by Anya Apavatjrut, Claire Goursaud, Katia Jaffres-Runser, Cristina Comaniciu, Jean-Marie Gorce

    Diversity is a powerful means to increase the transmission performance of wireless communications. For the case of fountain codes relaying, it has been shown previously that introducing diversity is also beneficial since it counteracts transmission losses on the channel. Instead of simply hop-by-hop forwarding information, each sensor node diversifies the information flow using XOR combinations of stored packets. This approach has been shown to be efficient for random linear fountain codes. However, random linear codes exhibit high decoding complexity. In this paper, we propose diversity increased relaying strategies for the more realistic and lower complexity Luby Transform code in a linear network. Results are provided herein for a linear network assuming uniform imperfect channel states.


  • Efficient cooperative detection for wireless sentinel networks

    by Qiong Shi, Cristina Comaniciu

    In this paper, we propose an energy efficient cooperative detection scheme for wireless sentinel networks monitoring for surreptitious wireless communications. Our solution exploits the multiuser diversity in the network by proposing a cooperative randomized monitoring approach for individual sentinel nodes, that leads to more efficient resource expenditure and good detection and intruder localization performance for the network. A game theoretic framework is proposed to analyze the achievable operating point for the system. Simulation results show very good detection performance together with increased localization accuracy and energy efficient operation.

  • Open Access

    A Multiobjective Optimization Framework for Routing in Wireless Ad Hoc Networks

    by Katia Jaffrès-Runser, Cristina Comaniciu, Jean-Marie Gorce

    Wireless ad hoc networks are seldom characterized by one single performance metric, yet the current literature lacks a flexible framework to assist in characterizing the design tradeoffs in such networks. In this work, we address this problem by proposing a new modeling framework for routing in ad hoc networks, which used in conjunction with metaheuristic multiobjective search algorithms, will result in a better understanding of network behavior and performance when multiple criteria are relevant. Our approach is to take a holistic view of the network that captures the cross-interactions among interference management techniques implemented at various layers of the protocol stack. The resulting framework is a complex multiobjective optimization problem that can be efficiently solved through existing multiobjective search techniques. In this contribution, we present the Pareto optimal sets for an example sensor network when delay, robustness and energy are considered. The aim of this paper is to present the framework and hence for conciseness purposes, the multiobjective optimization search is not developed herein.

  • Open Access

    Interference and Congestion Aware Gradient Broadcasting Routing for Wireless Sensor Networks

    by Katia Jaffrès-Runser, Cristina Comaniciu, Jean-Marie Gorce

    This paper addresses the problem of reliable transmission of data through a sensor network. We focus on networks rapidly deployed in harsh environments. For these networks, important design requirements are fast data transmission and rapid network setup, as well as minimized energy consumption for increased network lifetime. We propose a novel broadcasting solution that accounts for the interference impact and the congestion level of the channel, in order to improve robustness, energy consumption and delay performance, compared to a benchmark routing protocol, the GRAB algorithm. Three solutions are proposed: P-GRAB, a probabilistic routing algorithm for interference mitigation, U-GRAB, a utility-based algorithm that adjusts to real-time congestion and UP-GRAB, a combination of P-GRAB and U-GRAB. It is shown that P-GRAB provides the best performance for geometry-aware networks while the U-GRAB approach is the best option for unreliable and unstable networks.