Tu-Thach Quach

Energy Scaling Advantages of Resistive Memory Crossbar Based Computation and Its Application to Sparse Coding

The exponential increase in data over the last decade presents a significant challenge to analytics efforts that seek to process and interpret such data for various applications. Neural-inspired computing approaches are being developed in order to leverage the computational properties of the analog, low-power data processing observed in biological systems. Analog resistive memory crossbars can perform a parallel read or a vector-matrix multiplication as well as a parallel write or a rank-1 update with high computational efficiency. For an N × N crossbar, these two kernels can be O(N) more energy efficient than a conventional digital memory-based architecture. If the read operation is noise limited, the energy to read a column can be independent of the crossbar size (O(1)). These two kernels form the basis of many neuromorphic algorithms such as image, text, and speech recognition. For instance, these kernels can be applied to a neural sparse coding algorithm to give an O(N) reduction in energy for the entire algorithm when run with finite precision. Sparse coding is a rich problem with a host of applications including computer vision, object tracking, and more generally unsupervised learning.

Optimal Cover Estimation Methods and Steganographic Payload Location

Cover estimation is an important part of steganalysis and has many applications. One such application is steganographic payload location using residuals, which is effective when a large number of stego images are available. In the ideal case when the cover images are available, we show that the expected number of stego images needed to perfectly locate all load-carrying pixels is approximately the logarithm of the payload size. In more practical settings when the cover images are not available, the accuracy of payload location depends primarily on the chosen cover estimation method. We present optimal, linear runtime algorithms for finding the most likely cover estimate given the stego image and experimentally demonstrate that they can be used to locate payload on both least-significant bit (LSB) replacement and LSB matching stego images. The algorithms can be extended to higher order statistical models of cover images.

Cover estimation and payload location using Markov random fields

Payload location is an approach to find the message bits hidden in steganographic images, but not necessarily their logical order. Its success relies primarily on the accuracy of the underlying cover estimators and can be improved if more estimators are used. This paper presents an approach based on Markov random field to estimate the cover image given a stego image. It uses pairwise constraints to capture the natural two-dimensional statistics of cover images and forms a basis for more sophisticated models. Experimental results show that it is competitive against current state-of-the-art estimators and can locate payload embedded by simple LSB steganography and group-parity steganography. Furthermore, when combined with existing estimators, payload location accuracy improves significantly.

Extracting hidden messages in steganographic images

The eventual goal of steganalytic forensic is to extract the hidden messages embedded in steganographic images. A promising technique that addresses this problem partially is steganographic payload location, an approach to reveal the message bits, but not their logical order. It works by finding modified pixels, or residuals, as an artifact of the embedding process. This technique is successful against simple least-significant bit steganography and group-parity steganography. The actual messages, however, remain hidden as no logical order can be inferred from the located payload. This paper establishes an important result addressing this shortcoming: we show that the expected mean residuals contain enough information to logically order the located payload provided that the size of the payload in each stego image is not fixed. The located payload can be ordered as prescribed by the mean residuals to obtain the hidden messages without knowledge of the embedding key, exposing an inherent vulnerability in these embedding algorithms. Experimental results are provided to support our analysis.

A Diffusion Model for Maximizing Influence Spread in Large Networks

Influence spread is an important phenomenon that occurs in many social networks. Influence maximization is the corresponding problem of finding the most influential nodes in these networks. In this paper, we present a new influence diffusion model, based on pairwise factor graphs, that captures dependencies and directions of influence among neighboring nodes. We use an augmented belief propagation algorithm to efficiently compute influence spread on this model so that the direction of influence is preserved. Due to its simplicity, the model can be used on large graphs with high-degree nodes, making the influence maximization problem practical on large, real-world graphs. Using large Flixster and Epinions datasets, we provide experimental results showing that our model predictions match well with ground-truth influence spreads, far better than other techniques. Furthermore, we show that the influential nodes identified by our model achieve significantly higher influence spread compared to other popular models. The model parameters can easily be learned from basic, readily available training data. In the absence of training, our approach can still be used to identify influential seed nodes.

A model-based approach to finding tracks in SAR CCD images

Combining multiple synthetic aperture radar (SAR) images taken at different times of the same scene produces coherent change detection (CCD) images that can detect small surface changes such as tire tracks. The resulting CCD images can be used in an automated approach to identify and label tracks. Existing techniques have limited success due to the noisy nature of these CCD images. In particular, existing techniques require some user cues and can only trace a single track. This paper presents an approach to automatically identify and label multiple tracks in CCD images. We use an explicit objective function that utilizes the Bayesian information criterion to find the simplest set of curves that explains the observed data. Experimental results show that it is capable of identifying tracks under various scenes and can correctly declare when no tracks are present.

The energy scaling advantages of RRAM crossbars

As transistors start to approach fundamental limits and Moores law slows down, new devices and architectures are needed to enable continued performance gains. New approaches based on RRAM (resistive random access memory) or memristor crossbars can enable the processing of large amounts of data. One of the most promising applications for RRAM crossbars is brain inspired or neuromorphic computing.

Convolutional networks for vehicle track segmentation

Abstract. Existing methods to detect vehicle tracks in coherent change detection images, a product of combining two synthetic aperture radar images taken at different times of the same scene, rely on simple and fast models to label track pixels. These models, however, are unable to capture natural track features, such as continuity and parallelism. More powerful but computationally expensive models can be used in offline settings. We present an approach that uses dilated convolutional networks consisting of a series of 3×3 convolutions to segment vehicle tracks. The design of our networks considers the fact that remote sensing applications tend to operate in low power and have limited training data. As a result, we aim for small and efficient networks that can be trained end-to-end to learn natural track features entirely from limited training data. We demonstrate that our six-layer network, trained on just 90 images, is computationally efficient and improves the F-score on a standard dataset to 0.992, up from 0.959 obtained by the current state-of-the-art method.