Neural Network Layer Algebra: A Framework to Measure Capacity and Compression in Deep Learning.

We present a new framework to measure the intrinsic properties of (deep) neural networks. While we focus on convolutional networks, our framework can be extrapolated to any network architecture. In particular, we evaluate two network properties, namely, capacity, which is related to expressivity, and compression, which is related to learnability. Both these properties depend only on the network structure and are independent of the network parameters. To this end, we propose two metrics: the first one, called layer complexity, captures the architectural complexity of any network layer; and, the second one, called layer intrinsic power, encodes how data are compressed along the network. The metrics are based on the concept of layer algebra, which is also introduced in this article. This concept is based on the idea that the global properties depend on the network topology, and the leaf nodes of any neural network can be approximated using local transfer functions, thereby allowing a simple computation of the global metrics. We show that our global complexity metric can be calculated and represented more conveniently than the widely used Vapnik-Chervonenkis (VC) dimension. We also compare the properties of various state-of-the-art architectures using our metrics and use the properties to analyze their accuracy on benchmark image classification datasets.

Identifying characteristics that impact motor carrier safety using Bayesian networks.

PROBLEM STATEMENT: In the U.S., a safety rating is assigned to each motor carrier based on data obtained from the Motor Carrier Management Information System (MCMIS) and an on-site investigation. While researchers have identified variables associated with the safety ratings, the specific direction of the relationships are not necessarily clear. OBJECTIVE: The objective of this study is to identify those relationships involved in the safety ratings of interstate motor carriers, the largest users of the U.S. transportation network. METHOD: Bayesian networks are used to learn these relationships from data obtained from MCMIS for a 6-year period (2007-2012). RESULTS: Our study shows that safety rating assignment is a complex process with only a subset of the variables having statistically significant relationship with safety rating. They include driver out-of-service violations, weight violations, traffic violations, fleet size, total employed drivers, and passenger & general carrier indicators. APPLICATION: The findings have both immediate implications and long term benefits. The immediate implications relate to better identification of unsafe motor carriers, and the long term benefits pertain to policies and crash countermeasures that can enhance carrier safety.

Predicting interstate motor carrier crash rate level using classification models.

Ensuring safe operations of large commercial vehicles (motor carriers) remains an important challenge, particularly in the United States. While the federal regulatory agency has instituted a compliance review-based rating method to encourage carriers to improve their safety levels, concerns have been expressed regarding the effectiveness of the current ratings. In this paper, we consider a crash rate level (high, medium, and low) rather than a compliance review-based rating (satisfactory, conditional satisfactory, and unsatisfactory). We demonstrate an automated way of predicting the crash rate levels for each carrier using three different classification models (Artificial Neural Network, Classification and Regression Tree (CART), and Support Vector Machine) and three separate variable selection methods (Empirical Evidence, Multiple Factor Analysis, Garson's algorithm). The predicted crash rate levels (high, low) are compared to the assigned levels based on the current safety rating method. The results indicate the feasibility of crash rate level as an effective measure of carrier safety, with CART having the best performance.