Dorit Baras

Reinforcement Learning, Spike-Time-Dependent Plasticity, and the BCM Rule

Learning agents, whether natural or artificial, must update their internal parameters in order to improve their behavior over time. In reinforcement learning, this plasticity is influenced by an environmental signal, termed a reward, that directs the changes in appropriate directions. We apply a recently introduced policy learning algorithm from machine learning to networks of spiking neurons and derive a spike-time-dependent plasticity rule that ensures convergence to a local optimum of the expected average reward. The approach is applicable to a broad class of neuronal models, including the Hodgkin-Huxley model. We demonstrate the effectiveness of the derived rule in several toy problems. Finally, through statistical analysis, we show that the synaptic plasticity rule established is closely related to the widely used BCM rule, for which good biological evidence exists.

K-Means with Large and Noisy Constraint Sets

We focus on the problem of clustering with soft instance-level constraints. Recently, the CVQE algorithm was proposed in this context. It modifies the objective function of traditional K-means to include penalties for violated constraints. CVQE was shown to efficiently produce high-quality clustering of UCI data. In this work, we examine the properties of CVQE and propose a modification that results in a more intuitive objective function, with lower computational complexity. We present our extensive experimentation, which provides insight into CVQE and shows that our new variant can dramatically improve clustering quality while reducing run time. We show its superiority in a large-scale surveillance scenario with noisy constraints.

Automatic Boosting of Cross-Product Coverage Using Bayesian Networks

Closing the feedback loop from coverage data to the stimuli generator is one of the main challenges in the verification process. Typically, verification engineers with deep domain knowledge manually prepare a set of stimuli generation directives for that purpose. Bayesian networks based CDG (coverage directed generation) systems have been successfully used to assist the process by automatically closing this feedback loop. However, constructing these CDG systems requires manual effort and a certain amount of domain knowledge from a machine learning specialist. We propose a new method that boosts coverage at early stages of the verification process with minimal effort, namely a fully automatic construction of a CDG system that requires no domain knowledge. Experimental results on a real-life cross-product coverage model demonstrate the efficiency of the proposed method.

Analysis of advanced meter infrastructure data of water consumption in apartment buildings

We present our experience of using machine learning techniques over data originating from advanced meter infrastructure (AMI) systems for water consumption in a medium-size city. We focus on two new use cases that are of special importance to city authorities. One use case is the automatic identification of malfunctioning meters, with a focus on distinguishing them from legitimate non-consumption such as during periods when the household residents are on vacation. The other use case is the identification of leaks or theft in the unmetered common areas of apartment buildings. These two use cases are highly important to city authorities both because of the lost revenue they imply and because of the hassle to the residents in cases of delayed identification. Both cases are inherently complex to analyze and require advanced data mining techniques in order to achieve high levels of correct identification. Our results provide for faster and more accurate detection of malfunctioning meters as well as leaks in the common areas. This results in significant tangible value to the authorities in terms of increase in technician efficiency and a decrease in the amount of wasted, non-revenue, water.

Automatic boosting of cross-product coverage using Bayesian networks

Closing the feedback loop from coverage data to the stimuli generator is one of the main challenges in the verification process. Typically, verification engineers with deep domain knowledge manually prepare a set of stimuli generation directives for that purpose. Bayesian networks based CDG (coverage directed generation) systems have been successfully used to assist the process by automatically closing this feedback loop. However, constructing these CDG systems requires manual effort and a certain amount of domain knowledge from a machine learning specialist. We propose a new method that boosts coverage in the early stages of the verification process with minimal effort, namely a fully automatic construction of a CDG system that requires no domain knowledge. Experimental results on a real-life cross-product coverage model demonstrate the efficiency of the proposed method.