As the field of paralinguistic affect recognition has become more mature, many researchers have shifted their approach from a single channel of affect manifestation to a multi-modal one in developing their affect recognition systems. In the spirit continuing this trend in multi-modal work, our work utilizes paralinguistic features of speech and contextual knowledge. Through our human study, we found that contextual knowledge had positive impact on a human’s affect recognition ability when combined with paralinguistic features of speech. In this research, we propose a novel architecture called Context-Based Paralinguistic Affect Recognition System (CxBPARS) that combines the traditional paralinguistic affect recognition approach using classification algorithms and the contextual knowledge related to the emotion elicitors and their environment. By combining the results of an Ada-Boost classifier and contextual modeling, we achieved an improvement in affect recognition accuracy from 29.5% (context free) to 53.0% (context dependent).

In this paper, we report experimental results from augmenting Recurrent Neural Networks (RNN) with multi-step-ahead strategies for traffic speed prediction. For multi-step-ahead time series forecasting, researchers have applied MIMO, recursive, and direct strategies to machine learning methods in other domains. We applied the recursive and direct strategies to the GRU networks for predicting multi-step-ahead traffic speed and compared the prediction errors with the GRU network without the strategies (i.e. MIMO strategy). Based on the results from the experiments, we found that the direct strategy and the MIMO strategy produce models with smaller error metrics as compared to the recursive strategy. The direct strategy is computationally very expensive, thus MIMO strategy i.e. the GRU architecture without any strategy, is our preferred recommendation.

Writing style is a combination of consistent decisions at different levels of language production including lexical, syntactic, and structural associated to a specific author (or author groups). While lexical-based models have been widely explored in style-based text classification, relying on content makes the model less scalable when dealing with heterogeneous data comprised of various topics. On the other hand, syntactic models which are content-independent, are more robust against topic variance. In this paper, we introduce a syntactic recurrent neural network to encode the syntactic patterns of a document in a hierarchical structure. The model first learns the syntactic representation of sentences from the sequence of part-of-speech tags. Subsequently, the syntactic representations of sentences are aggregated into document representation using recurrent neural networks. Our experimental results on PAN 2012 dataset for authorship attribution task shows that syntactic recurrent neural network outperforms the lexical model with the identical architecture by approximately 14\% in terms of accuracy.

Modeling player engagement is a key challenge in games. However, the gameplay signatures of engaged players can be highly context-sensitive, varying based on where the game is used or what population of players is using it. Traditionally, models of player engagement are investigated in a particular context, and it is unclear how effectively these models generalize to other settings and populations. In this work, we investigate a Bayesian hierarchical linear model for multi-task learning to devise a model of player engagement from a pair of datasets that were gathered in two complementary contexts: a Classroom Study with middle school students and a Laboratory Study with undergraduate students. Both groups of players used similar versions of Crystal Island, an educational interactive narrative game for science learning. Results indicate that the Bayesian hierarchical model outperforms both pooled and context-specific models in cross-validation measures of predicting player motivation from in-game behaviors, particularly for the smaller Classroom Study group. Further, we find that the posterior distributions of model parameters indicate that the coefficient for a measure of gameplay performance significantly differs between groups. Drawing upon their capacity to share information across groups, hierarchical Bayesian methods provide an effective approach for modeling player engagement with data from similar, but different, contexts.

We examine the effects of inter-agent variation on the ability of a decentralized multi-agent system (MAS) to self-organize in response to dynamically changing task demands. In decentralized biological systems, inter-agent variation as minor as noise has been observed to improve a system's ability to redistribute agent resources in response to external stimuli. We compare the performance of two MAS consisting of agents with and without noisy sensors on a cooperative tracking problem and examine the effects of inter-agent variation on agent behaviors and how those behaviors affect system performance. Results show that small variations in how individual agents respond to stimuli can lead to more accurate and stable allocation of agent resources.

Detection of abnormal behavior is the catalyst for many applications that seek to react to deviations from behavioral expectations. However, this is often difficult to do when direct communication with the performer is impractical. Therefore, we propose to create models of normal human performance and then compare their performance to a human's actual behavior. Any detected deviations can be then used to determine what condition(s) could possibly be influencing the deviant behavior. We build the models of human behavior through machine learning from observation; more specifically, we employ the Genetic Context Learning algorithm to create models of normal car driving behaviors of different humans with and without ADHD (Attention Deficit Hyperactivity Disorder). We use a car simulator for our studies to eliminate risk to our test subjects and to other drivers. Our results show that different driving situations have varying utility in abnormal behavior detection. Learning from Observation was successful in building models to be applied to abnormal behavior detection.

Healthcare informatics focuses on the efficient and effective acquisition, management, and use of information in healthcare. Advancing health informatics has been declared a grand challenge by the National Academy of Engineering and is a major area of emphasis for agencies such as the Centers for Medicare and Medicaid Services. As such, it has been identified as an area of national need. Sample uses of AI in health informatics includes expert systems for decision support, machine learning and data mining to discover patterns across patients, image analysis to assist in diagnosis, and natural language processing to extract information from free text medical documents. The areas of interest for this track include healthcare decision support, medical image processing, machine learning and data mining in healthcare, processing and managing patient records, syndromic surveillance, drug discovery, and personalization of clinical care.

Interactive training environments typically include feedback mechanisms designed to help trainees improve their performance through either guided or self-reflection. In this context, trainees are candidate teachers who need to hone their social skills as well as other pedagogical skills for their future classroom. We chose an avatar-mediated interactive virtual training system–TeachLivE–as the basic research environment to investigate the motions and embodiment of the trainees. Using tracking sensors, and customized improvements for existing gesture recognition utilities, we created a gesture database and employed it for the implementation of our real-time gesture recognition and feedback application. We also investigated multiple methods of feedback provision, including visual and haptics. The results from the conducted user studies and user evaluation surveys indicate the positive impact of the proposed feedback applications and informed body language. In this paper, we describe the context in which the utilities have been developed, the importance of recognizing nonverbal communication in the teaching context, the means of providing automated feedback associated with nonverbal messaging, and the preliminary studies developed to inform the research.

Robots assisting the disabled or elderly must perform complex manipulation tasks and must adapt to the home environment and preferences of their user. Learning from demonstration is a promising choice, that would allow the non-technical user to teach the robot different tasks. However, collecting demonstrations in the home environment of a disabled user is time consuming, disruptive to the comfort of the user, and presents safety challenges. It would be desirable to perform the demonstrations in a virtual environment. In this paper we describe a solution to the challenging problem of behavior transfer from virtual demonstration to a physical robot. The virtual demonstrations are used to train a deep neural network based controller, which is using a Long Short Term Memory (LSTM) recurrent neural network to generate trajectories. The training process uses a Mixture Density Network (MDN) to calculate an error signal suitable for the multimodal nature of demonstrations. The controller learned in the virtual environment is transferred to a physical robot (a Rethink Robotics Baxter). An off-the-shelf vision component is used to substitute for geometric knowledge available in the simulation and an inverse kinematics module is used to allow the Baxter to enact the trajectory. Our experimental studies validate the three contributions of the paper: (1) the controller learned from virtual demonstrations can be used to successfully perform the manipulation tasks on a physical robot, (2) the LSTM+MDN architectural choice outperforms other choices, such as the use of feedforward networks and mean-squared error based training signals and (3) allowing imperfect demonstrations in the training set also allows the controller to learn how to correct its manipulation mistakes.

Reports on the 2017 AAAI Spring Symposium Series