IDK Cascades are a framework used to improve time usage within classification tasks. A cascade is comprised of a series of classification models with increasing levels of complexity. A prediction occurs when a single model's computed confidence is above a certain threshold; otherwise, an IDK classification is made, which leads to the next model in the series considering the input. A recent improvement to this scheme is the idea of skipping a model in the case where the prior model's confidence was extremely low. However, this method has inherent issues when predicting whether to skip. This paper improves on the method of skipping certain models by using a Random Forest model trained on multiple parameters, resulting in smart skipping. Unlike the previous threshold skipping approach, the Random Forest's is able to more accurately predict when a skip should occur, leading to larger time savings. Ultimately, we are able to obtain an IDK Cascade utilizing Random Forest skipping that is more computationally efficient than threshold skipping across three datasets.

When routing vehicles through a traffic network, routes are generated to reduce the costs of vehicles traveling through the network with the typical cost function being derived from travel time. What is less considered is the cost derived from vehicle destination-arrival deadlines, while also considering vehicle travel time. In this paper, we explore the extension of a cost function via a vehicle interaction function aimed at reduced vehicle deadline misses by penalizing vehicles that interact with vehicles that miss deadlines. We modify a Dynamic Traffic Assignment implementation in the traffic simulation Simulation Of Urban MObility (SUMO) and test our method on a real-world network. We show that our method is able to outperform Dynamic User Equilibrium and Dynamic System Optimal formulations in tardiness.

In an effort to reduce the runtime of classification algorithms, IDK (I Don’t Know) cascades have been presented as an alternative to current classification models. These structures comprise of a “cascade” of classifiers that only categorizes an input if a classifier outputs a confidence level that exceeds a predetermined threshold. If it does not, it outputs the class “I don’t know” and moves on to the next classifier in the cascade. However, these IDK cascades often reach their worst-case execution time, where classification is only completed by the final classifier. This paper aims to improve the static structure employed by these cascades, deploying a dynamic IDK framework that skips certain classifiers upon meeting specific conditions. A probabilistic analysis run on a previously validated optimal IDK cascade helped us identify these specific conditions, using the outputted confidence level of the first classifier to dictate if a cascade should run its middle classifiers. Ultimately, we generated a dynamic IDK cascade that ran up to 17% faster than its static counterpart.

Effective communication between a clinician and their patient is critical for delivering healthcare maximizing outcomes. Unfortunately, traditional communication training approaches that use human standardized patients and expert coaches are difficult to scale. Here, we present the development and validation of a scalable, easily accessible, digital tool known as the Standardized Online Patient for Health Interaction Education (SOPHIE) for practicing and receiving feedback on doctor-patient communication skills. SOPHIE was validated by conducting an experiment with 30 medical professionals. We found that participants who underwent SOPHIE condition performed significantly better than the control in overall communication, aggregate scores, empowering the patient, and showing empathy (p < 0.05 in all cases). The results presented in this paper provide early evidence of scalable and accessible virtual humans, chatbot technologies, and automated feedback generation being transformative for medical communication training by supplementing existing resources.

A scaled smart city consists of a combination of infrastructure and vehicular agents working in concert to direct traffic throughout the vehicular network at a smaller-thanlife scale. In this paper, we outline our plans for routing to satisfy arrival deadlines, where vehicles are routed with the primary objective of getting somewhere on time. We consider vehicle routing through a traffic sub-network, using a centralized scheme as a guiding traffic assignment agent. We introduce our preliminary implementation of a routing algorithm built on the Selfless Traffic Routing (STR) model and Dynamic User Equilibrium (DUE) to show the viability of such a scheme on a traffic network. We present our experimental results from running this scheme on a real-world traffic network.