Publications

Abstract not provided.

The prevalence effect is the observation that, in visual search tasks as the signal (target) to noise (non-target) ratio becomes smaller, humans are more likely to miss the target when it does occur. Studied extensively in the basic literature [e.g., 1, 2], this effect has implications for real-world settings such as security guards monitoring physical facilities for attacks. Importantly, what seems to drive the effect is the development of a response bias based on learned sensitivity to the statistical likelihood of a target [e.g., 3-5]. This paper presents results from two experiments aimed at understanding how the target prevalence impacts the ability for individuals to detect a target on the 1,000th trial of a series of 1000 trials. The first experiment employed the traditional prevalence effect paradigm. This paradigm involves search for a perfect capital letter T amidst imperfect Ts. In a between-subjects design, our subjects experienced target prevalence rates of 50/50, 1/10, 1/100, or 1/1000. In all conditions, the final trial was always a target. The second (ongoing) experiment replicates this design using a notional physical facility in a mod/sim environment. This simulation enables triggering different intrusion detection sensors by simulated characters and events (e.g., people, animals, weather). In this experiment, subjects viewed 1000 “alarm” events and were asked to characterize each as either a nuisance alarm (e.g., set off by an animal) or an attack. As with the basic visual search study, the final trial was always an attack.

File fragment classification is an important step in the task of file carving in digital forensics. In file carving, files must be reconstructed based on their content as a result of their fragmented storage on disk or in memory. Existing methods for classification of file fragments typically use hand-engineered features, such as byte histograms or entropy measures. In this paper, we propose an approach using sparse coding that enables automated feature extraction. Sparse coding, or sparse dictionary learning, is an unsupervised learning algorithm, and is capable of extracting features based simply on how well those features can be used to reconstruct the original data. With respect to file fragments, we learn sparse dictionaries for n-grams, continuous sequences of bytes, of different sizes. These dictionaries may then be used to estimate n-gram frequencies for a given file fragment, but for significantly larger n-gram sizes than are typically found in existing methods which suffer from combinatorial explosion. To demonstrate the capability of our sparse coding approach, we used the resulting features to train standard classifiers, such as support vector machines over multiple file types. Experimentally, we achieved significantly better classification results with respect to existing methods, especially when the features were used in supplement to existing hand-engineered features.

Abstract not provided.

Abstract not provided.

Abstract not provided.

While existing work in neural interfaces is largely geared toward the restoration of lost function in amputees or victims of neurological injuries, similar technology may also facilitate augmentation of healthy subjects. One example is the potential to learn a new, unnatural sense through a neural interface. The use of neural interfaces in healthy subjects would require an even greater level of safety and convenience than in disabled subjects, including reliable, robust bidirectional implants with highly-portable components outside the skin. We present our progress to date in the development of a bidirectional neural interface system intended for completely untethered use. The system consists of a wireless stimulating and recording peripheral nerve implant powered by a rechargeable battery, and a wearable package that communicates wirelessly both with the implant and with a computer or a network of independent sensor nodes. Once validated, such a system could permit the exploration of increasingly realistic use of neural interfaces both for restoration and for augmentation. © 2007 IEEE.

Abstract not provided.