A5066795279- Reinforcement Learning in Robotics
- Explainable Artificial Intelligence (XAI)
- Adversarial Robustness in Machine Learning
- Machine Learning and Data Classification
- Robot Manipulation and Learning
- Data Stream Mining Techniques
- Topic Modeling
- Ethics and Social Impacts of AI
- AI-based Problem Solving and Planning
- Statistical and Computational Modeling
- Software Engineering Research
- Machine Learning in Healthcare
- Social Robot Interaction and HRI
- Information and Cyber Security
- Speech and dialogue systems
- Machine Learning and Algorithms
- Fuzzy Logic and Control Systems
- Bayesian Modeling and Causal Inference
- Evolutionary Algorithms and Applications
- Evolutionary Game Theory and Cooperation
- Neural Networks and Applications
- Anomaly Detection Techniques and Applications
- Botulinum Toxin and Related Neurological Disorders
- Artificial Intelligence in Healthcare and Education
- Temporomandibular Joint Disorders
Carnegie Mellon University
2019-2024
University of Maryland, Baltimore County
2019
University of Naples Federico II
2007
University of Milan
2007
Though deep reinforcement learning has led to breakthroughs in many difficult domains, these successes have required an ever-increasing number of samples. As state-of-the-art (RL) systems require exponentially increasing samples, their development is restricted a continually shrinking segment the AI community. Likewise, cannot be applied real-world problems, where environment samples are expensive. Resolution limitations requires new, sample-efficient methods. To facilitate research this...
Explainable reinforcement learning (XRL) is an emerging subfield of explainable machine that has attracted considerable attention in recent years. The goal XRL to elucidate the decision-making process agents sequential settings. In this survey, we propose a novel taxonomy for organizing literature prioritizes RL setting. We overview techniques according taxonomy. point out gaps literature, which use motivate and outline roadmap future work.
Unilateral posterior crossbite has been considered as a risk factor for temporomandibular joint clicking, with conflicting findings. The aim of this study was to investigate possible association between unilateral and disk displacement reduction, by means survey carried out in young adolescents recruited from three schools. sample included 1291 participants (708 males 583 females) mean age 12.3 yrs (range, 10.1–16.1 yrs), who underwent an orthodontic functional examination performed two...
As domestic service robots become more common and widespread, they must be programmed to efficiently accomplish tasks while aligning their actions with relevant norms. The first step equip normative reasoning competence is understanding the norms that people apply behavior of in specific social contexts. To end, we conducted an online survey Chinese United States participants which asked them select preferred action a robot should take number scenarios. paper makes multiple contributions....
We aim to understand how people assess human likeness in navigation produced by and artificially intelligent (AI) agents a video game. To this end, we propose novel AI agent with the goal of generating more human-like behavior. collect hundreds crowd-sourced assessments comparing human-likeness behavior generated our baseline human-generated Our proposed passes Turing Test, while do not. By passing mean that judges could not quantitatively distinguish between videos person an navigating....
Current work in explainable reinforcement learning generally produces policies the form of a decision tree over state space. Such can be used for formal safety verification, agent behavior prediction, and manual inspection important features. However, existing approaches fit after training or use custom procedure which is not compatible with new techniques, such as those neural networks. To address this limitation, we propose novel Markov Decision Process (MDP) type policies: Iterative...
To facilitate research in the direction of sample efficient reinforcement learning, we held MineRL Competition on Sample Efficient Reinforcement Learning Using Human Priors at Thirty-third Conference Neural Information Processing Systems (NeurIPS 2019). The primary goal this competition was to promote development algorithms that use human demonstrations alongside learning reduce number samples needed solve complex, hierarchical, and sparse environments. We describe competition, outlining...
We introduce an algorithm for model-based hierarchical reinforcement learning to acquire self-contained transition and reward models suitable probabilistic planning at multiple levels of abstraction. call this framework Planning with Abstract Learned Models (PALM). By representing subtasks symbolically using a new formal structure, the lifted abstract Markov decision process (L-AMDP), PALM learns that are independent modular. Through our experiments, we show how integrates execution,...
To facilitate research in the direction of fine-tuning foundation models from human feedback, we held MineRL BASALT Competition on Fine-Tuning Human Feedback at NeurIPS 2022. The challenge asks teams to compete develop algorithms solve tasks with hard-to-specify reward functions Minecraft. Through this competition, aimed promote development that use feedback as channels learn desired behavior. We describe competition and provide an overview top solutions. conclude by discussing impact future...
Multi-agent reinforcement learning (MARL) methods often suffer from high sample complexity, limiting their use in real-world problems where data is sparse or expensive to collect. Although latent-variable world models have been employed address this issue by generating abundant synthetic for MARL training, most of these cannot encode vital global information available during training into latent states, which hampers efficiency. The few exceptions that incorporate assume centralized...
The last decade has seen a significant increase of interest in deep learning research, with many public successes that have demonstrated its potential. As such, these systems are now being incorporated into commercial products. With this comes an additional challenge: how can we build AI solve tasks where there is not crisp, well-defined specification? While multiple solutions been proposed, competition focus on one particular: from human feedback. Rather than training using predefined...
Reinforcement learning competitions advance the field by providing appropriate scope and support to develop solutions toward a specific problem. To promote development of more broadly applicable methods, organizers need enforce use general techniques, sample-efficient reproducibility results. While beneficial for research community, these restrictions come at cost -- increased difficulty. If barrier entry is too high, many potential participants are demoralized. With this in mind, we hosted...
The goal of this paper is to understand how people assess human-likeness in human- and AI-generated behavior. To end, we present a qualitative study hundreds crowd-sourced assessments behavior 3D video game navigation task. In particular, focus on an AI agent that has passed Turing Test, the sense human judges were not able reliably distinguish between videos navigating quantitative level. Our insights shine light characteristics consider as human-like. Understanding these key first step for...
Reinforcement learning (RL) has recently shown promise in predicting Alzheimer's disease (AD) progression due to its unique ability model domain knowledge. However, it is not clear which RL algorithms are well-suited for this task. Furthermore, these methods inherently explainable, limiting their applicability real-world clinical scenarios. Our work addresses two important questions. Using a causal, interpretable of AD, we first compare the performance four contemporary brain cognition over...
Understanding the mechanisms behind decisions taken by large foundation models in sequential decision making tasks is critical to ensuring that such systems operate transparently and safely. In this work, we perform exploratory analysis on Video PreTraining (VPT) Minecraft playing agent, one of largest open-source vision-based agents. We aim illuminate its reasoning applying various interpretability techniques. First, analyze attention mechanism while agent solves training task - crafting a...
Recent advances in reinforcement learning (RL) have predominantly leveraged neural network-based policies for decision-making, yet these models often lack interpretability, posing challenges stakeholder comprehension and trust. Concept bottleneck offer an interpretable alternative by integrating human-understandable concepts into networks. However, a significant limitation prior work is the assumption that human annotations are readily available during training, necessitating continuous...
To encourage the development of methods with reproducible and robust training behavior, we propose a challenge paradigm where competitors are evaluated directly on performance their learning procedures rather than pre-trained agents. Since competition organizers re-train proposed in controlled setting they can guarantee reproducibility, -- by retraining submissions using held-out test set help ensure generalization past environments which were trained.
The MineRL BASALT competition has served to catalyze advances in learning from human feedback through four hard-to-specify tasks Minecraft, such as create and photograph a waterfall. Given the completion of two years competitions, we offer community formalized benchmark Evaluation Demonstrations Dataset (BEDD), which serves resource for algorithm development performance assessment. BEDD consists collection 26 million image-action pairs nearly 14,000 videos players completing Minecraft. It...