Motivation: Many researchers with domain expertise are unable to easily apply machine learning to their bioinformatics data due to a lack of machine learning and/or coding expertise. Methods that have been proposed thus far to automate machine learning mostly require programming experience as well as expert knowledge to tune and apply the algorithms correctly. Here, we study a method of automating biomedical data science using a web-based platform that uses AI to recommend model choices and conduct experiments. We have two goals in mind: first, to make it easy to construct sophisticated models of biomedical processes; and second, to provide a fully automated AI agent that can choose and conduct promising experiments for the user, based on the user's experiments as well as prior knowledge. To validate this framework, we experiment with hundreds of classification problems, comparing to state-of-the-art, automated approaches. Finally, we use this tool to develop predictive models of septic shock in critical care patients. Results: We find that matrix factorization-based recommendation systems outperform meta-learning methods for automating machine learning. This result mirrors the results of earlier recommender systems research in other domains. The proposed AI is competitive with state-of-the-art automated machine learning methods in terms of choosing optimal algorithm configurations for datasets. In our application to prediction of septic shock, the AI-driven analysis produces a competent machine learning model (AUROC 0.85 +/- 0.02) that performs on par with state-of-the-art deep learning results for this task, with much less computational effort.
Modern power grids are experiencing grand challenges caused by the stochastic and dynamic nature of growing renewable energy and demand response. Traditional theoretical assumptions and operational rules may be violated, which are difficult to be adapted by existing control systems due to the lack of computational power and accurate grid models for use in real time, leading to growing concerns in the secure and economic operation of the power grid. Existing operational control actions are typically determined offline, which are less optimized. This paper presents a novel paradigm, Grid Mind, for autonomous grid operational controls using deep reinforcement learning. The proposed AI agent for voltage control can learn its control policy through interactions with massive offline simulations, and adapts its behavior to new changes including not only load/generation variations but also topological changes. A properly trained agent is tested on the IEEE 14-bus system with tens of thousands of scenarios, and promising performance is demonstrated in applying autonomous voltage controls for secure grid operation.
Visual dialog (VisDial) is a task which requires an AI agent to answer a series of questions grounded in an image. Unlike in visual question answering (VQA), the series of questions should be able to capture a temporal context from a dialog history and exploit visually-grounded information. A problem called visual reference resolution involves these challenges, requiring the agent to resolve ambiguous references in a given question and find the references in a given image. In this paper, we propose Dual Attention Networks (DAN) for visual reference resolution. DAN consists of two kinds of attention networks, REFER and FIND. Specifically, REFER module learns latent relationships between a given question and a dialog history by employing a self-attention mechanism. FIND module takes image features and reference-aware representations (i.e., the output of REFER module) as input, and performs visual grounding via bottom-up attention mechanism. We qualitatively and quantitatively evaluate our model on VisDial v1.0 and v0.9 datasets, showing that DAN outperforms the previous state-of-the-art model by a significant margin.
Human behaviors are regularized by a variety of norms or regulations, either to maintain orders or to enhance social welfare. If artificially intelligent (AI) agents make decisions on behalf of human beings, we would hope they can also follow established regulations while interacting with humans or other AI agents. However, it is possible that an AI agent can opt to disobey the regulations (being defective) for self-interests. In this paper, we aim to answer the following question: Consider a multi-agent decentralized environment. Agents make decisions in complete isolation of other agents. Each agent knows the state of its own MDP and its own actions but it does not know the states and the actions taken by other players. There is a set of regulations for all agents to follow. Although most agents are benign and will comply to regulations but not all agents are compliant at first, can we develop a framework such that it is in the self-interest of non-compliant agents to comply after all?. We first introduce the problem as Regulation Enforcement and formulate it using reinforcement learning and game theory under the scenario where agents make decisions in complete isolation of other agents. We then propose a solution based on the key idea that although we could not alter how defective agents choose to behave, we can, however, leverage the aggregated power of compliant agents to boycott the defective ones. We conducted simulated experiments on two scenarios: Replenishing Resource Management Dilemma and Diminishing Reward Shaping Enforcement, using deep multi-agent reinforcement learning algorithms. We further use empirical game-theoretic analysis to show that the method alters the resulting empirical payoff matrices in a way that promotes compliance (making mutual compliant a Nash Equilibrium).
Many modern machine learning approaches require vast amounts of training data to learn new concepts; conversely, human learning often requires few examples--sometimes only one--from which the learner can abstract structural concepts. We present a novel approach to introducing new spatial structures to an AI agent, combining deep learning over qualitative spatial relations with various heuristic search algorithms. The agent extracts spatial relations from a sparse set of noisy examples of block-based structures, and trains convolutional and sequential models of those relation sets. To create novel examples of similar structures, the agent begins placing blocks on a virtual table, uses a CNN to predict the most similar complete example structure after each placement, an LSTM to predict the most likely set of remaining moves needed to complete it, and recommends one using heuristic search. We verify that the agent learned the concept by observing its virtual block-building activities, wherein it ranks each potential subsequent action toward building its learned concept. We empirically assess this approach with human participants' ratings of the block structures. Initial results and qualitative evaluations of structures generated by the trained agent show where it has generalized concepts from the training data, which heuristics perform best within the search space, and how we might improve learning and execution.
Item response theory (IRT) can be applied to the analysis of the evaluation of results from AI benchmarks. The two-parameter IRT model provides two indicators (difficulty and discrimination) on the side of the item (or AI problem) while only one indicator (ability) on the side of the respondent (or AI agent). In this paper we analyse how to make this set of indicators dual, by adding a fourth indicator, generality, on the side of the respondent. Generality is meant to be dual to discrimination, and it is based on difficulty. Namely, generality is defined as a new metric that evaluates whether an agent is consistently good at easy problems and bad at difficult ones. With the addition of generality, we see that this set of four key indicators can give us more insight on the results of AI benchmarks. In particular, we explore two popular benchmarks in AI, the Arcade Learning Environment (Atari 2600 games) and the General Video Game AI competition. We provide some guidelines to estimate and interpret these indicators for other AI benchmarks and competitions.
Starcraft II (SC2) is widely considered as the most challenging Real Time Strategy (RTS) game. The underlying challenges include a large observation space, a huge (continuous and infinite) action space, partial observations, simultaneous move for all players, and long horizon delayed rewards for local decisions. To push the frontier of AI research, Deepmind and Blizzard jointly developed the StarCraft II Learning Environment (SC2LE) as a testbench of complex decision making systems. SC2LE provides a few mini games such as MoveToBeacon, CollectMineralShards, and DefeatRoaches, where some AI agents have achieved the performance level of human professional players. However, for full games, the current AI agents are still far from achieving human professional level performance. To bridge this gap, we present two full game AI agents in this paper - the AI agent TStarBot1 is based on deep reinforcement learning over a flat action structure, and the AI agent TStarBot2 is based on hard-coded rules over a hierarchical action structure. Both TStarBot1 and TStarBot2 are able to defeat the built-in AI agents from level 1 to level 10 in a full game (1v1 Zerg-vs-Zerg game on the AbyssalReef map), noting that level 8, level 9, and level 10 are cheating agents with unfair advantages such as full vision on the whole map and resource harvest boosting. To the best of our knowledge, this is the first public work to investigate AI agents that can defeat the built-in AI in the StarCraft II full game.
Existing visual explanation generating agents learn to fluently justify a class prediction. However, they may mention visual attributes which reflect a strong class prior, although the evidence may not actually be in the image. This is particularly concerning as ultimately such agents fail in building trust with human users. To overcome this limitation, we propose a phrase-critic model to refine generated candidate explanations augmented with flipped phrases which we use as negative examples while training. At inference time, our phrase-critic model takes an image and a candidate explanation as input and outputs a score indicating how well the candidate explanation is grounded in the image. Our explainable AI agent is capable of providing counter arguments for an alternative prediction, i.e. counterfactuals, along with explanations that justify the correct classification decisions. Our model improves the textual explanation quality of fine-grained classification decisions on the CUB dataset by mentioning phrases that are grounded in the image. Moreover, on the FOIL tasks, our agent detects when there is a mistake in the sentence, grounds the incorrect phrase and corrects it significantly better than other models.
Artificial Intelligence is becoming instrumental in a variety of applications. Games serve as a good breeding ground for trying and testing these algorithms in a sandbox with simpler constraints in comparison to real life. In this project, we aim to develop an AI agent that can solve the classical Japanese game of Sokoban using various algorithms and heuristics and compare their performances through standard metrics.
Natural language explanations of deep neural network decisions provide an intuitive way for a AI agent to articulate a reasoning process. Current textual explanations learn to discuss class discriminative features in an image. However, it is also helpful to understand which attributes might change a classification decision if present in an image (e.g., "This is not a Scarlet Tanager because it does not have black wings.") We call such textual explanations counterfactual explanations, and propose an intuitive method to generate counterfactual explanations by inspecting which evidence in an input is missing, but might contribute to a different classification decision if present in the image. To demonstrate our method we consider a fine-grained image classification task in which we take as input an image and a counterfactual class and output text which explains why the image does not belong to a counterfactual class. We then analyze our generated counterfactual explanations both qualitatively and quantitatively using proposed automatic metrics.
It is hypothesized by some thinkers that benign looking AI objectives may result in powerful AI drives that may pose an existential risk to human society. We analyze this scenario and find the underlying assumptions to be unlikely. We examine the alternative scenario of what happens when universal goals that are not human-centric are used for designing AI agents. We follow a design approach that tries to exclude malevolent motivations from AI agents, however, we see that objectives that seem benevolent may pose significant risk. We consider the following meta-rules: preserve and pervade life and culture, maximize the number of free minds, maximize intelligence, maximize wisdom, maximize energy production, behave like human, seek pleasure, accelerate evolution, survive, maximize control, and maximize capital. We also discuss various solution approaches for benevolent behavior including selfless goals, hybrid designs, Darwinism, universal constraints, semi-autonomy, and generalization of robot laws. A "prime directive" for AI may help in formulating an encompassing constraint for avoiding malicious behavior. We hypothesize that social instincts for autonomous robots may be effective such as attachment learning. We mention multiple beneficial scenarios for an advanced semi-autonomous AGI agent in the near future including space exploration, automation of industries, state functions, and cities. We conclude that a beneficial AI agent with intelligence beyond human-level is possible and has many practical use cases.
There is a growing focus on how to design safe artificial intelligent (AI) agents. As systems become more complex, poorly specified goals or control mechanisms may cause AI agents to engage in unwanted and harmful outcomes. Thus it is necessary to design AI agents that follow initial programming intentions as the program grows in complexity. How to specify these initial intentions has also been an obstacle to designing safe AI agents. Finally, there is a need for the AI agent to have redundant safety mechanisms to ensure that any programming errors do not cascade into major problems. Humans are autonomous intelligent agents that have avoided these problems and the present manuscript argues that by understanding human self-regulation and goal setting, we may be better able to design safe AI agents. Some general principles of human self-regulation are outlined and specific guidance for AI design is given.
This article provides the first survey of computational models of emotion in reinforcement learning (RL) agents. The survey focuses on agent/robot emotions, and mostly ignores human user emotions. Emotions are recognized as functional in decision-making by influencing motivation and action selection. Therefore, computational emotion models are usually grounded in the agent's decision making architecture, of which RL is an important subclass. Studying emotions in RL-based agents is useful for three research fields. For machine learning (ML) researchers, emotion models may improve learning efficiency. For the interactive ML and human-robot interaction (HRI) community, emotions can communicate state and enhance user investment. Lastly, it allows affective modelling (AM) researchers to investigate their emotion theories in a successful AI agent class. This survey provides background on emotion theory and RL. It systematically addresses 1) from what underlying dimensions (e.g., homeostasis, appraisal) emotions can be derived and how these can be modelled in RL-agents, 2) what types of emotions have been derived from these dimensions, and 3) how these emotions may either influence the learning efficiency of the agent or be useful as social signals. We also systematically compare evaluation criteria, and draw connections to important RL sub-domains like (intrinsic) motivation and model-based RL. In short, this survey provides both a practical overview for engineers wanting to implement emotions in their RL agents, and identifies challenges and directions for future emotion-RL research.
The domain of text-based adventure games has been recently established as a new challenge of creating the agent that is both able to understand natural language, and acts intelligently in text-described environments. In this paper, we present our approach to tackle the problem. Our agent, named Golovin, takes advantage of the limited game domain. We use genre-related corpora (including fantasy books and decompiled games) to create language models suitable to this domain. Moreover, we embed mechanisms that allow us to specify, and separately handle, important tasks as fighting opponents, managing inventory, and navigating on the game map. We validated usefulness of these mechanisms, measuring agent's performance on the set of 50 interactive fiction games. Finally, we show that our agent plays on a level comparable to the winner of the last year Text-Based Adventure AI Competition.
Question-answering (QA) on video contents is a significant challenge for achieving human-level intelligence as it involves both vision and language in real-world settings. Here we demonstrate the possibility of an AI agent performing video story QA by learning from a large amount of cartoon videos. We develop a video-story learning model, i.e. Deep Embedded Memory Networks (DEMN), to reconstruct stories from a joint scene-dialogue video stream using a latent embedding space of observed data. The video stories are stored in a long-term memory component. For a given question, an LSTM-based attention model uses the long-term memory to recall the best question-story-answer triplet by focusing on specific words containing key information. We trained the DEMN on a novel QA dataset of children's cartoon video series, Pororo. The dataset contains 16,066 scene-dialogue pairs of 20.5-hour videos, 27,328 fine-grained sentences for scene description, and 8,913 story-related QA pairs. Our experimental results show that the DEMN outperforms other QA models. This is mainly due to 1) the reconstruction of video stories in a scene-dialogue combined form that utilize the latent embedding and 2) attention. DEMN also achieved state-of-the-art results on the MovieQA benchmark.
We introduce the task of Visual Dialog, which requires an AI agent to hold a meaningful dialog with humans in natural, conversational language about visual content. Specifically, given an image, a dialog history, and a question about the image, the agent has to ground the question in image, infer context from history, and answer the question accurately. Visual Dialog is disentangled enough from a specific downstream task so as to serve as a general test of machine intelligence, while being grounded in vision enough to allow objective evaluation of individual responses and benchmark progress. We develop a novel two-person chat data-collection protocol to curate a large-scale Visual Dialog dataset (VisDial). VisDial v0.9 has been released and contains 1 dialog with 10 question-answer pairs on ~120k images from COCO, with a total of ~1.2M dialog question-answer pairs. We introduce a family of neural encoder-decoder models for Visual Dialog with 3 encoders -- Late Fusion, Hierarchical Recurrent Encoder and Memory Network -- and 2 decoders (generative and discriminative), which outperform a number of sophisticated baselines. We propose a retrieval-based evaluation protocol for Visual Dialog where the AI agent is asked to sort a set of candidate answers and evaluated on metrics such as mean-reciprocal-rank of human response. We quantify gap between machine and human performance on the Visual Dialog task via human studies. Putting it all together, we demonstrate the first 'visual chatbot'! Our dataset, code, trained models and visual chatbot are available on https://visualdialog.org
We present Octopus, an AI agent to jointly balance three conflicting task objectives on a micro-crowdsourcing marketplace - the quality of work, total cost incurred, and time to completion. Previous control agents have mostly focused on cost-quality, or cost-time tradeoffs, but not on directly controlling all three in concert. A naive formulation of three-objective optimization is intractable; Octopus takes a hierarchical POMDP approach, with three different components responsible for setting the pay per task, selecting the next task, and controlling task-level quality. We demonstrate that Octopus significantly outperforms existing state-of-the-art approaches on real experiments. We also deploy Octopus on Amazon Mechanical Turk, showing its ability to manage tasks in a real-world dynamic setting.
Reinforcement Learning AI commonly uses reward/penalty signals that are objective and explicit in an environment -- e.g. game score, completion time, etc. -- in order to learn the optimal strategy for task performance. However, Human-AI interaction for such AI agents should include additional reinforcement that is implicit and subjective -- e.g. human preferences for certain AI behavior -- in order to adapt the AI behavior to idiosyncratic human preferences. Such adaptations would mirror naturally occurring processes that increase trust and comfort during social interactions. Here, we show how a hybrid brain-computer-interface (hBCI), which detects an individual's level of interest in objects/events in a virtual environment, can be used to adapt the behavior of a Deep Reinforcement Learning AI agent that is controlling a virtual autonomous vehicle. Specifically, we show that the AI learns a driving strategy that maintains a safe distance from a lead vehicle, and most novelly, preferentially slows the vehicle when the human passengers of the vehicle encounter objects of interest. This adaptation affords an additional 20\% viewing time for subjectively interesting objects. This is the first demonstration of how an hBCI can be used to provide implicit reinforcement to an AI agent in a way that incorporates user preferences into the control system.
Generating natural language descriptions of images is an important capability for a robot or other visual-intelligence driven AI agent that may need to communicate with human users about what it is seeing. Such image captioning methods are typically trained by maximising the likelihood of ground-truth annotated caption given the image. While simple and easy to implement, this approach does not directly maximise the language quality metrics we care about such as CIDEr. In this paper we investigate training image captioning methods based on actor-critic reinforcement learning in order to directly optimise non-differentiable quality metrics of interest. By formulating a per-token advantage and value computation strategy in this novel reinforcement learning based captioning model, we show that it is possible to achieve the state of the art performance on the widely used MSCOCO benchmark.
In the quest towards general artificial intelligence (AI), researchers have explored developing loss functions that act as intrinsic motivators in the absence of external rewards. This paper argues that such research has overlooked an important and useful intrinsic motivator: social interaction. We posit that making an AI agent aware of implicit social feedback from humans can allow for faster learning of more generalizable and useful representations, and could potentially impact AI safety. We collect social feedback in the form of facial expression reactions to samples from Sketch RNN, an LSTM-based variational autoencoder (VAE) designed to produce sketch drawings. We use a Latent Constraints GAN (LC-GAN) to learn from the facial feedback of a small group of viewers, by optimizing the model to produce sketches that it predicts will lead to more positive facial expressions. We show in multiple independent evaluations that the model trained with facial feedback produced sketches that are more highly rated, and induce significantly more positive facial expressions. Thus, we establish that implicit social feedback can improve the output of a deep learning model.