3IA PhD/Postdoc Seminar #13

Informative labels in Semi-Supervised Learning

Abstract: 
In semi-supervised learning, we have access to features but the outcome variable is missing for a part of the data. In real life, although the amount of data available is often huge, labeling the data is costly and time-consuming. It is particularly true for image data sets: images are available in large quantities on image banks but they are most of the time unlabeled. It is therefore necessary to ask experts to label them. In this context, people are more inclined to label images of some classes which are easy to recognize. The unlabeled data are thus informative missing values, because the unavailability of the labels depends on their values themselves. Typically, the goal of semi-supervised learning is to learn predictive models using all the data (labeled and unlabeled ones). However, classical methods lead to biased estimates if the missing values are informative. We aim at designing new semi-supervised algorithms that handle missing labels, possibly informative.

Data driven traffic management by Macroscopic models

Abstract:  
We propose a calibration framework for traffic reconstruction and prediction, exploiting the loop detector data set provided by the Minnesota Department of Transportation (MnDOT). We validate the results comparing the error metrics of both first and second order macroscopic traffic flow models, showing that the usage of the traffic flow model leads to reasonable prediction accuracy.

Macroscopic traffic flow models have been employed for decades to describe the spatio-temporal evolution of aggregate traffic quantities such as density and mean velocity. Classically, macroscopic traffic models are calibrated either by fitting the so-called fundamental diagram (i.e., the density-flow or density-speed mapping described by the model flux function) or by minimizing some error measure of the simulation output. The calibration can be done against either data provided by loop detectors at fixed locations or trajectory data. Following Kennedy-O’Hagan [1], in [3] we introduce a bias term to better account for possible discrepancies between the mathematical models and reality; this bias term is modeled by a Gaussian process. A Dynamic Time Warping algorithm is then applied to predict future traffic conditions at loop detector locations and sparse time points. These serve as initial data to simulate the traffic conditions at a finer scale, leading to reasonable and more accurate travel time predictions.

Open discussion on the two contributions