Total: 11
While recent advances in deep reinforcement learning have allowed autonomous learning agents to succeed at a variety of complex tasks, existing algorithms generally require a lot oftraining data. One way to increase the speed at which agent sare able to learn to perform tasks is by leveraging the input of human trainers. Although such input can take many forms, real-time, scalar-valued feedback is especially useful in situations where it proves difficult or impossible for humans to provide expert demonstrations. Previous approaches have shown the usefulness of human input provided in this fashion (e.g., the TAMER framework), but they have thus far not considered high-dimensional state spaces or employed the use of deep learning. In this paper, we do both: we propose DeepTAMER, an extension of the TAMER framework that leverages the representational power of deep neural networks inorder to learn complex tasks in just a short amount of time with a human trainer. We demonstrate Deep TAMER’s success by using it and just 15 minutes of human-provided feedback to train an agent that performs better than humans on the Atari game of Bowling - a task that has proven difficult for even state-of-the-art reinforcement learning methods.
Research has shown that a person's financial success is more dependent on the ability to deal with people than on professional knowledge. Sage advice, such as "if you can't say something nice, don't say anything at all" and principles articulated in Carnegie's classic "How to Win Friends and Influence People," offer trusted rules-of-thumb for how people can successfully deal with each other. However, alternative philosophies for dealing with people have also emerged. The success of an AI system is likewise contingent on its ability to win friends and influence people. In this paper, we study how AI systems should be designed to win friends and influence people in repeated games with cheap talk (RGCTs). We create several algorithms for playing RGCTs by combining existing behavioral strategies (what the AI does) with signaling strategies (what the AI says) derived from several competing philosophies. Via user study, we evaluate these algorithms in four RGCTs. Our results suggest sufficient properties for AIs to win friends and influence people in RGCTs.
Assessing the veracity of claims made on the Internet is an important, challenging, and timely problem. While automated fact-checking models have potential to help people better assess what they read, we argue such models must be explainable, accurate, and fast to be useful in practice; while prediction accuracy is clearly important, model transparency is critical in order for users to trust the system and integrate their own knowledge with model predictions. To achieve this, we propose a novel probabilistic graphical model (PGM) which combines machine learning with crowd annotations. Nodes in our model correspond to claim veracity, article stance regarding claims, reputation of news sources, and annotator reliabilities. We introduce a fast variational method for parameter estimation. Evaluation across two real-world datasets and three scenarios shows that: (1) joint modeling of sources, claims and crowd annotators in a PGM improves the predictive performance and interpretability for predicting claim veracity; and (2) our variational inference method achieves scalably fast parameter estimation, with only modest degradation in performance compared to Gibbs sampling. Regarding model transparency, we designed and deployed a prototype fact-checker Web tool, including a visual interface for explaining model predictions. Results of a small user study indicate that model explanations improve user satisfaction and trust in model predictions. We share our web demo, model source code, and the 13K crowd labels we collected.
Agents that can learn new tasks through interactive instruction can utilize goal information to search for and learn flexible policies. This approach can be resilient to variations in initial conditions or issues that arise during execution. However, if a task is not easily formulated as achieving a goal or if the agent lacks sufficient domain knowledge for planning, other methods are required. We present a hybrid approach to interactive task learning that can learn both goal-oriented and procedural tasks, and mixtures of the two, from human natural language instruction. We describe this approach, go through two examples of learning tasks, and outline the space of tasks that the system can learn. We show that our approach can learn a variety of goal-oriented and procedural tasks from a single example and is robust to different amounts of domain knowledge.
Volunteers who help with online crowdsourcing such as citizen science tasks typically make only a few contributions before exiting. We propose a computational approach for increasing users' engagement in such settings that is based on optimizing policies for displaying motivational messages to users. The approach, which we refer to as Trajectory Corrected Intervention (TCI), reasons about the tradeoff between the long-term influence of engagement messages on participants' contributions and the potential risk of disrupting their current work. We combine model-based reinforcement learning with off-line policy evaluation to generate intervention policies, without relying on a fixed representation of the domain. TCI works iteratively to learn the best representation from a set of random intervention trials and to generate candidate intervention policies. It is able to refine selected policies off-line by exploiting the fact that users can only be interrupted once per session.We implemented TCI in the wild with Galaxy Zoo, one of the largest citizen science platforms on the web. We found that TCI was able to outperform the state-of-the-art intervention policy for this domain, and significantly increased the contributions of thousands of users. This work demonstrates the benefit of combining traditional AI planning with off-line policy methods to generate intelligent intervention strategies.
We propose a scheme for training a computerized agent to perform complex human tasks such as highway steering. The scheme is designed to follow a natural learning process whereby a human instructor teaches a computerized trainee. It enables leveraging the weak supervision abilities of a (human) instructor, who, while unable to perform well herself at the required task, can provide coherent and learnable instantaneous reward signals to the computerized trainee. The learning process consists of three supervised elements followed by reinforcement learning. The supervised learning stages are: (i) supervised imitation learning; (ii) supervised reward induction; and (iii) supervised safety module construction. We implemented this scheme using deep convolutional networks and applied it to successfully create a computerized agent capable of autonomous highway steering over the well-known racing game Assetto Corsa. We demonstrate that the use of all components is essential to effectively carry out reinforcement learning of the steering task using vision alone, without access to a driving simulator internals, and operating in wall-clock time.
We introduce a novel model-agnostic system that explains the behavior of complex models with high-precision rules called anchors, representing local, "sufficient" conditions for predictions. We propose an algorithm to efficiently compute these explanations for any black-box model with high-probability guarantees. We demonstrate the flexibility of anchors by explaining a myriad of different models for different domains and tasks. In a user study, we show that anchors enable users to predict how a model would behave on unseen instances with less effort and higher precision, as compared to existing linear explanations or no explanations.
By capturing statistical patterns in large corpora, machine learning has enabled significant advances in natural language processing, including in machine translation, question answering, and sentiment analysis. However, for agents to intelligently interact with humans, simply capturing the statistical patterns is insufficient. In this paper we investigate if, and how, grounded compositional language can emerge as a means to achieve goals in multi-agent populations. Towards this end, we propose a multi-agent learning environment and learning methods that bring about emergence of a basic compositional language. This language is represented as streams of abstract discrete symbols uttered by agents over time, but nonetheless has a coherent structure that possesses a defined vocabulary and syntax. We also observe emergence of non-verbal communication such as pointing and guiding when language communication is unavailable.
A classifier’s low confidence in prediction is often indicative of whether its prediction will be wrong; in this case, inputs are called known unknowns. In contrast, unknown unknowns (UUs) are inputs on which a classifier makes a high confidence mistake. Identifying UUs is especially important in safety-critical domains like medicine (diagnosis) and law (recidivism prediction). Previous work by Lakkaraju et al. (2017) on identifying unknown unknowns assumes that the utility of each revealed UU is independent of the others, rather than considering the set holistically. While this assumption yields an efficient discovery algorithm, we argue that it produces an incomplete understanding of the classifier’s limitations. In response, this paper proposes a new class of utility models that rewards how well the discovered UUs cover (or "explain") a sample distribution of expected queries. Although choosing an optimal cover is intractable, even if the UUs were known, our utility model is monotone submodular, affording a greedy discovery strategy. Experimental results on four datasets show that our method outperforms bandit-based approaches and achieves within 60.9% utility of an omniscient, tractable upper bound.
Multi-label classification is crucial to several practical applications including document categorization, video tagging, targeted advertising etc. Training a multi-label classifier requires a large amount of labeled data which is often unavailable or scarce. Labeled data is then acquired by consulting multiple labelers---both human and machine. Inspired by ensemble methods, our premise is that labels inferred with high consensus among labelers, might be closer to the ground truth. We propose strategies based on interaction and active learning to obtain higher quality labels that potentially lead to greater consensus. We propose a novel formulation that aims to collectively optimize the cost of labeling, labeler reliability, label-label correlation and inter-labeler consensus. Evaluation on data labeled by multiple labelers (both human and machine) shows that our consensus output is closer to the ground truth when compared to the "majority" baseline. We present illustrative cases where it even improves over the existing ground truth. We also present active learning strategies to leverage our consensus model in interactive learning settings. Experiments on several real-world datasets (publicly available) demonstrate the efficacy of our approach in achieving promising classification results with fewer labeled data.
Building large-scale, globally consistent maps is a challenging problem, made more difficult in environments with limited access, sparse features, or when using data collected by novice users. For such scenarios, where state-of-the-art mapping algorithms produce globally inconsistent maps, we introduce a systematic approach to incorporating sparse human corrections, which we term Human-in-the-Loop Simultaneous Localization and Mapping (HitL-SLAM). Given an initial factor graph for pose graph SLAM, HitL-SLAM accepts approximate, potentially erroneous, and rank-deficient human input, infers the intended correction via expectation maximization (EM), back-propagates the extracted corrections over the pose graph, and finally jointly optimizes the factor graph including the human inputs as human correction factor terms, to yield globally consistent large-scale maps. We thus contribute an EM formulation for inferring potentially rank-deficient human corrections to mapping, and human correction factor extensions to the factor graphs for pose graph SLAM that result in a principled approach to joint optimization of the pose graph while simultaneously accounting for multiple forms of human correction. We present empirical results showing the effectiveness of HitL-SLAM at generating globally accurate and consistent maps even when given poor initial estimates of the map.