Deep learning models, though having achieved great success in many different fields over the past years, are usually data-hungry, fail to perform well on unseen samples, and lack interpretability. Different kinds of prior knowledge often exists in the target domain, and their use can alleviate the deficiencies with deep learning. To better mimic the behavior of human brains, different advanced methods have been proposed to identify domain knowledge and integrate it into deep models for data-efficient, generalizable, and interpretable deep learning, which we refer to as knowledge-augmented deep learning (KADL). In this survey, we define the concept of KADL and introduce its three major tasks, i.e., knowledge identification, knowledge representation, and knowledge integration. Different from existing surveys that are focused on a specific type of knowledge, we provide a broad and complete taxonomy of domain knowledge and its representations. Based on our taxonomy, we provide a systematic review of existing techniques, different from existing works that survey integration approaches agnostic to the taxonomy of knowledge. This survey subsumes existing works and offers a bird's-eye view of research in the general area of KADL. The thorough and critical reviews of numerous papers help not only understand current progress but also identify future directions for the research on KADL.

Probabilistic inference can be tackled by minimizing a variational free energy through message passing. To improve performance, neural networks are adopted for message computation. Neural message learning is heuristic and requires strong guidance to perform well. In this work, we propose a {\em theory-guided message passing neural network} (TMPNN) for probabilistic inference. Inspired by existing work, we consider a generalized Bethe free energy which allows for a learnable variational assumption. Instead of using a black-box neural network for message computation, we utilize a general message equation and introduce a symbolic message function with semantically meaningful parameters. The analytically derived symbolic message function is seamlessly integrated into the MPNN framework, giving rise to the proposed TMPNN. TMPNN is trained using algorithmic supervision without requiring exact inference results. Leveraging the theory-guided symbolic function, TMPNN offers strengthened theoretical guarantees compared to conventional heuristic neural models. It presents a novel contribution by demonstrating its applicability to both MAP and marginal inference tasks, outperforming SOTAs in both cases. Furthermore, TMPNN provides improved generalizability across various graph structures and exhibits enhanced data efficiency.

Kinjal Basu, Keerthiram Murugesan, Subhajit Chaudhury, Murray Campbell, Kartik Talamadupula, Tim Klinger. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

We introduce VisualHandicaps, a novel benchmark environment for the systematic analysis of interactive text-based reinforcement learning (TBRL) agents by providing visual handicaps. Unlike previous TBRL environments, which focus on providing additional textual information to measure agent understanding of sequential natural language information, VisualHandicaps seeks to improve the generalization ability of RL agents using varying details of maps and textual information, allowing for the study and demonstration of robust planning and self-localization. We provide automatically generated variations and difficulty levels in our environment and show that an agent using our systematic visual handicaps along with textual observation generally outperforms previous methods (that use only textual handicaps) in terms of success rate and the number of steps required to reach the goal. We also provide a detailed analysis of each handicap, which we believe to be important findings for driving future improvements in RL agents on text-based applications.

Abstract Constraining the actions of AI systems is one promising way to ensure that these systems behave in a way that is morally acceptable to humans. But constraints alone come with drawbacks as in many AI systems, they are not flexible. If these constraints are too rigid, they can preclude actions that are actually acceptable in certain, contextual situations. Humans, on the other hand, can often decide when a simple and seemingly inflexible rule should actually be overridden based on the context. In this paper, we empirically investigate the way humans make these contextual moral judgements, with the goal of building AI systems that understand when to follow and when to override constraints. We propose a novel and general preference-based graphical model that captures a modification of standard dual process theories of moral judgment. We then detail the design, implementation, and results of a study of human participants who judge whether it is acceptable to break a well-established rule: no cutting in line. We then develop an instance of our model and compare its performance to that of standard machine learning approaches on the task of predicting the behavior of human participants in the study, showing that our preference-based approach more accurately captures the judgments of human decision-makers. It also provides a flexible method to model the relationship between variables for moral decision-making tasks that can be generalized to other settings.

Existing AU detection algorithms are mainly based on appearance information extracted from 2D images, and well-established facial biomechanics that governs 3D facial skin deformation is rarely considered. In this paper, we propose a biomechanics-guided AU detection approach, where facial muscle activation forces are modelled and are employed to predict AU activation. Specifically, our model consists of two branches: 3D physics branch and 2D image branch. In 3D physics branch, we first derive the Euler-Lagrange equation governing facial deformation. The Euler-Lagrange equation represented as an ordinary differential equation (ODE) is embedded into a differentiable ODE solver. Muscle activation forces together with other physics parameters are firstly regressed, and then are utilized to simulate 3D deformation by solving the ODE. By leveraging facial biomechanics, we obtain physically plausible facial muscle activation forces. 2D image branch compensates 3D physics branch by employing additional appearance information from 2D images. Both estimated forces and appearance features are employed for AU detection. The proposed approach achieves competitive AU detection performance on two benchmark datasets. Furthermore, by leveraging biomechanics, our approach achieves outstanding performance with reduced training data.

While natural language systems continue improving, they are still imperfect. If a user has a better understanding of how a system works, they may be able to better accomplish their goals even in imperfect systems. We explored whether explanations can support effective authoring of natural language utterances and how those explanations impact users’ mental models in the context of a natural language system that generates small programs. Through an online study (n=252), we compared two main types of explanations: 1) system-focused, which provide information about how the system processes utterances and matches terms to a knowledge base, and 2) social, which provide information about how other users have successfully interacted with the system. Our results indicate that providing social suggestions of terms to add to an utterance helped users to repair and generate correct flows more than system-focused explanations or social recommendations of words to modify. We also found that participants commonly understood some mechanisms of the natural language system, such as the matching of terms to a knowledge base, but they often lacked other critical knowledge, such as how the system handled structuring and ordering. Based on these findings, we make design recommendations for supporting interactions with and understanding of natural language systems.

Web applications and services are increasingly important in a distributed internet filled with diverse cloud services and applications, each of which enable the completion of narrowly defined tasks. Given the explosion in the scale and diversity of such services, their composition and integration for achieving complex user goals remains a challenging task for end-users and requires a lot of development effort when specified by hand. We present a demonstration of the Goal Oriented Flow Assistant (GOFA) system, which provides a natural language solution to generate workflows for application integration. Our tool is built on a three-step pipeline: it first uses Abstract Meaning Representation (AMR) to parse utterances; it then uses a knowledge graph to validate candidates; and finally uses an AI planner to compose the candidate flow. We provide a video demonstration of the deployed system as part of our submission.

Text-based games (TBGs) have become a popular proving ground for the demonstration of learning-based agents that make decisions in quasi real-world settings. The crux of the problem for a reinforcement learning agent in such TBGs is identifying the objects in the world, and those objects' relations with that world. While the recent use of text-based resources for increasing an agent's knowledge and improving its generalization have shown promise, we posit in this paper that there is much yet to be learned from visual representations of these same worlds. Specifically, we propose to retrieve images that represent specific instances of text observations from the world and train our agents on such images. This improves the agent's overall understanding of the game scene and objects' relationships to the world around them, and the variety of visual representations on offer allow the agent to generate a better generalization of a relationship. We show that incorporating such images improves the performance of agents in various TBG settings.

What does it mean for a generative AI model to be explainable? The emergent discipline of explainable AI (XAI) has made great strides in helping people understand discriminative models. Less attention has been paid to generative models that produce artifacts, rather than decisions, as output. Meanwhile, generative AI (GenAI) technologies are maturing and being applied to application domains such as software engineering. Using scenario-based design and question-driven XAI design approaches, we explore users’ explainability needs for GenAI in three software engineering use cases: natural language to code, code translation, and code auto-completion. We conducted 9 workshops with 43 software engineers in which real examples from state-of-the-art generative AI models were used to elicit users’ explainability needs. Drawing from prior work, we also propose 4 types of XAI features for GenAI for code and gathered additional design ideas from participants. Our work explores explainability needs for GenAI for code and demonstrates how human-centered approaches can drive the technical development of XAI in novel domains.

Generative machine learning models have recently been applied to source code, for use cases including translating code between programming languages, creating documentation from code, and auto-completing methods. Yet, state-of-the-art models often produce code that is erroneous or incomplete. In a controlled study with 32 software engineers, we examined whether such imperfect outputs are helpful in the context of Java-to-Python code translation. When aided by the outputs of a code translation model, participants produced code with fewer errors than when working alone. We also examined how the quality and quantity of AI translations affected the work process and quality of outcomes, and observed that providing multiple translations had a larger impact on the translation process than varying the quality of provided translations. Our results tell a complex, nuanced story about the benefits of generative code models and the challenges software engineers face when working with their outputs. Our work motivates the need for intelligent user interfaces that help software engineers effectively work with generative code models in order to understand and evaluate their outputs and achieve superior outcomes to working alone.

Maximum-A-Posteriori (MAP) inference is a fundamental task in probabilistic inference and belief propagation (BP) is a widely used algorithm for MAP inference. Though BP has been applied successfully to many different fields, it offers no performance guarantee and often performs poorly on loopy graphs. To improve the performance on loopy graphs and to scale up to large graphs, we propose a variational message passing neural network (V-MPNN), where we leverage both the power of neural networks in modeling complex functions and the well-established algorithmic theories on variational belief propagation. Instead of relying on a hand-crafted variational assumption, we propose a neural-augmented free energy where a general variational distribution is parameterized through a neural network. A message passing neural network is utilized for the minimization of neural-augmented free energy. Training of the MPNN is thus guided by neural-augmented free energy, without requiring exact MAP configurations as annotations. We empirically demonstrate the effectiveness of the proposed V-MPNN by comparing against both state-of-the-art training-free methods and training-based methods.

Knowledge graphs (KGs) are of great importance to many real world applications, but they generally suffer from incomplete information in the form of missing relations between entities. Knowledge graph completion (also known as relation prediction) is the task of inferring missing facts given existing ones. Most of the existing work is proposed by maximizing the likelihood of observed instance-level triples. Not much attention, however, is paid to the ontological information, such as type information of entities and relations. In this work, we propose a type-augmented relation prediction (TaRP) method, where we apply both the type information and instance-level information for the relation prediction. In particular, type information and instance-level information are encoded as prior probabilities and likelihoods of relations respectively, and are combined by following the Bayes' rule. Our proposed TaRP method achieves significantly better performance than state-of-the-art methods on four benchmark datasets: FB15K, FB15K-237, YAGO26K-906, and DB111K-174. In addition, we show that the TaRP achieves the significantly improved data efficiency. More importantly, the type information extracted from a specific dataset can generalize well to different datasets through the proposed TaRP model.

Text-based games have emerged as an important test-bed for Reinforcement Learning (RL) research, requiring RL agents to combine grounded language understanding with sequential decision making. In this paper, we examine the problem of infusing RL agents with commonsense knowledge. Such knowledge would allow agents to efficiently act in the world by pruning out implausible actions, and to perform look-ahead planning to determine how current actions might affect future world states. We design a new text-based gaming environment called TextWorld Commonsense (TWC) for training and evaluating RL agents with a specific kind of commonsense knowledge about objects, their attributes, and affordances. We also introduce several baseline RL agents which track the sequential context and dynamically retrieve the relevant commonsense knowledge from ConceptNet. We show that agents which incorporate commonsense knowledge in TWC perform better, while acting more efficiently. We conduct user-studies to estimate human performance on TWC and show that there is ample room for future improvement.

Generative models have become adept at producing artifacts such as images, videos, and prose at human-like levels of proficiency. New generative techniques, such as unsupervised neural machine translation (NMT), have recently been applied to the task of generating source code, translating it from one programming language to another. The artifacts produced in this way may contain imperfections, such as compilation or logical errors. We examine the extent to which software engineers would tolerate such imperfections and explore ways to aid the detection and correction of those errors. Using a design scenario approach, we interviewed 11 software engineers to understand their reactions to the use of an NMT model in the context of application modernization, focusing on the task of translating source code from one language to another. Our three-stage scenario sparked discussions about the utility and desirability of working with an imperfect AI system, how acceptance of that system’s outputs would be established, and future opportunities for generative AI in application modernization. Our study highlights how UI features such as confidence highlighting and alternate translations help software engineers work with and better understand generative NMT models.

Anshuman Mishra, Dhruvesh Patel, Aparna Vijayakumar, Xiang Lorraine Li, Pavan Kapanipathi, Kartik Talamadupula. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Textual entailment is a fundamental task in natural language processing. Most approaches for solving this problem use only the textual content present in training data. A few approaches have shown that information from external knowledge sources like knowledge graphs (KGs) can add value, in addition to the textual content, by providing background knowledge that may be critical for a task. However, the proposed models do not fully exploit the information in the usually large and noisy KGs, and it is not clear how it can be effectively encoded to be useful for entailment. We present an approach that complements text-based entailment models with information from KGs by (1) using Personalized PageRank to generate contextual subgraphs with reduced noise and (2) encoding these subgraphs using graph convolutional networks to capture the structural and semantic information in KGs. We evaluate our approach on multiple textual entailment datasets and show that the use of external knowledge helps the model to be robust and improves prediction accuracy. This is particularly evident in the challenging BreakingNLI dataset, where we see an absolute improvement of 5-20% over multiple text-based entailment models.

Subhajit Chaudhury, Daiki Kimura, Kartik Talamadupula, Michiaki Tatsubori, Asim Munawar, Ryuki Tachibana. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020.

Natural Language Inference (NLI) is fundamental to many Natural Language Processing (NLP) applications including semantic search and question answering. The NLI problem has gained significant attention due to the release of large scale, challenging datasets. Present approaches to the problem largely focus on learning-based methods that use only textual information in order to classify whether a given premise entails, contradicts, or is neutral with respect to a given hypothesis. Surprisingly, the use of methods based on structured knowledge – a central topic in artificial intelligence – has not received much attention vis-a-vis the NLI problem. While there are many open knowledge bases that contain various types of reasoning information, their use for NLI has not been well explored. To address this, we present a combination of techniques that harness external knowledge to improve performance on the NLI problem in the science questions domain. We present the results of applying our techniques on text, graph, and text-and-graph based models; and discuss the implications of using external knowledge to solve the NLI problem. Our model achieves close to state-of-the-art performance for NLI on the SciTail science questions dataset.

We demonstrate an embodied cognitive agent that helps scientists visualize and analyze exo-planets and their host stars. The prototype is situated in a room equipped with a large display, microphones, cameras, speakers, and pointing devices. Users communicate with the agent via speech, gestures, and combinations thereof, and it responds by displaying content and generating synthesized speech. Extensive use of context facilitates natural interaction with the agent.

In this demonstration, we report on the visualization capabilities of an Explainable AI Planning (XAIP) agent that can support human-in-the-loop decision-making. Imposing transparency and explainability requirements on such agents is crucial for establishing human trust and common ground with an end-to-end automated planning system. Visualizing the agent's internal decision making processes is a crucial step towards achieving this. This may include externalizing the "brain" of the agent: starting from its sensory inputs, to progressively higher order decisions made by it in order to drive its planning components. We demonstrate these functionalities in the context of a smart assistant in the Cognitive Environments Laboratory at IBM's T.J. Watson Research Center.

We introduce a new class of models called multiresolution recurrent neural networks, which explicitly model natural language generation at multiple levels of abstraction. The models extend the sequence-to-sequence framework to generate two parallel stochastic processes: a sequence of high-level coarse tokens, and a sequence of natural language words (e.g. sentences). The coarse sequences follow a latent stochastic process with a factorial representation, which helps the models generalize to new examples. The coarse sequences can also incorporate task-specific knowledge, when available. In our experiments, the coarse sequences are extracted using automatic procedures, which are designed to capture compositional structure and semantics. These procedures enable training the multiresolution recurrent neural networks by maximizing the exact joint log-likelihood over both sequences. We apply the models to dialogue response generation in the technical support domain and compare them with several competing models. The multiresolution recurrent neural networks outperform competing models by a substantial margin, achieving state-of-the-art results according to both a human evaluation study and automatic evaluation metrics. Furthermore, experiments show the proposed models generate more fluent, relevant and goal-oriented responses.

With the proliferation of technology, connected and interconnected devices (henceforth referred to as IoT) are fast becoming a viable option to automate the day-to-day interactions of users with their environment—be it manufacturing or home-care automation. However, with the explosion of IoT deployments we have observed in recent years, manually governing the interactions between humans-to-devices—and especially devices-to- devices—is an impractical task, if not an impossible task. This is because devices have their own obligations and prohibitions in context, and humans are not equip to maintain a bird’s-eye-view of the interaction space. Motivated by this observation, in this paper, we propose an end-to-end framework that (a) automatically dis- covers devices, and their associated services and capabilities w.r.t. an ontology; (b) supports representation of high-level—and expressive—user policies to govern the devices and services in the environment; (c) pro- vides efficient procedur es to refine and reason about policies to automate the management of interactions; and (d) delegates similar capable devices to fulfill the interactions, when conflicts occur. We then present our initial work in instrumenting the framework and discuss its details.

Recently there has been a lot of focus on human robot co-habitation issues that are often orthogonal to many aspects of human-robot teaming; e.g. on producing socially acceptable behaviors of robots and de-conflicting plans of robots and humans in shared environments. However, an interesting offshoot of these settings that has largely been overlooked is the problem of planning for serendipity - i.e. planning for stigmergic collaboration without explicit commitments on agents in co-habitation. In this paper we formalize this notion of planning for serendipity for the first time, and provide an Integer Programming based solution for this problem. Further, we illustrate the different modes of this planning technique on a typical Urban Search and Rescue scenario and show a real-life implementation of the ideas on the Nao Robot interacting with a human colleague.

One subclass of human computation applications are those directed at tasks that involve planning (e.g. tour planning) and scheduling (e.g. conference scheduling). Interestingly, work on these systems shows that even primitive forms of automated oversight on the human contributors helps in significantly improving the effectiveness of the humans/crowd. In this paper, we argue that the automated oversight used in these systems can be viewed as a primitive automated planner, and that there are several opportunities for more sophisticated automated planning in effectively steering the crowd. Straightforward adaptation of current planning technology is however hampered by the mismatch between the capabilities of human workers and automated planners. We identify and partially address two important challenges that need to be overcome before such adaptation of planning technology can occur: (i) interpreting inputs of the human workers (and the requester) and (ii) steering or critiquing plans produced by the human workers, armed only with incomplete domain and preference models. To these ends, we describe the implementation of AI-MIX, a tour plan generation system that uses automated checks and alerts to improve the quality of plans created by human workers; and present a preliminary evaluation of the effectiveness of steering provided by automated planning.

One of the more important aims of graduate artificial intelligence courses is to prepare graduate students to critically evaluate the current literature. The established approaches for this include either asking a student to present a paper in class, or to have the entire class read and discuss a paper. However, neither of these approaches presents incentives for student participation beyond the posting of a single summary or review. In this paper, we describe a class project that uses the popular Easychair conference management system as a pedagogical tool to enable engagement in the peer review process. We report on the deployment of this project in a mediumsized graduate AI class, and present the results of this deployment. We hope that the success of this project in engaging students in the peer review process can be used better train and bolster the future corps of AI reviewers.

Human computation applications that involve planning and scheduling are gaining popularity, and the existing literature on such systems shows that any automated oversight on human contributors improves the effectiveness of the crowd. In this paper, we present our ongoing work on the AI-MIX system, which is a first step towards using an automated planning and scheduling system in a crowdsourced planning application. In order to address the mismatch between the capabilities of the crowd and the automated planner, we identify two major challenges -- interpretation, and steering. We also present preliminary empirical results over the tour planning domain, and show how using an automated planner can help improve the quality of plans.

Beliefs play an important role in human-robot teaming scenarios, where the robots must reason about other agents' intentions and beliefs in order to inform their own plan generation process, and to successfully coordinate plans with the other agents. In this paper, we cast the evolving and complex structure of beliefs, and inference over them, as a planning and plan recognition problem. We use agent beliefs and intentions modeled in terms of predicates in order to create an automated planning problem instance, which is then used along with a known and complete domain model in order to predict the plan of the agent whose beliefs are being modeled. Information extracted from this predicted plan is used to inform the planning process of the modeling agent, to enable coordination. We also look at an extension of this problem to a plan recognition problem. We conclude by presenting an evaluation of our technique through a case study implemented on a real robot.

The increasing popularity of Twitter renders improved trust- worthiness and relevance assessment of tweets much more important for search. However, given the limitations on the size of tweets, it is hard to extract measures for ranking from the tweets? content alone. We present a novel ranking method called RAProp, which combines two orthogonal measures of relevance and trustworthiness of a tweet. The first, called Feature Score, measures the trustworthiness of the source of the tweet by extracting features from a 3-layer Twitter ecosystem consisting of users, tweets and webpages. The second measure, called agreement analysis, estimates the trustworthiness of the content of a tweet by analyzing whether the content is independently corroborated by other tweets. We view the candidate result set of tweets as the vertices of a graph, with the edges measuring the estimated agreement between each pair of tweets. The feature score is propagated over this agreement graph to compute the top-k tweets that have both trustworthy sources and independent corroboration. The evaluation of our method on 16 million tweets from the TREC 2011 Microblog Dataset shows that for top-30 precision, we achieve 53% better precision than the current best performing method on the data set, and an improvement of 300% over current Twitter Search.

Robots are currently being used in and developed for critical HRI applications such as search and rescue. In these scenarios, humans operating under changeable and high-stress conditions must communicate effectively with autonomous agents, necessitating that such agents be able to respond quickly and effectively to rapidly-changing conditions and expectations. We demonstrate a robot planner that is able to utilize new information, specifically information originating in spoken input produced by human operators.

In this paper, we present an integrated planning and robotic architecture that actively directs an agent engaged in an urban search and rescue (USAR) scenario. We describe three salient features that comprise the planning component of this system, namely (1) the ability to plan in a world open with respect to objects, (2) execution monitoring and replanning abilities, and (3) handling soft goals, and detail the interaction of these parts in representing and solving the USAR scenario at hand. We show that though insufficient in an individual capacity, the integration of this trio of features is sufficient to solve the scenario that we present. We test our system with an example problem that involves soft and hard goals, as well as goal deadlines and action costs, and show that the planner is capable of incorporating sensing actions and execution monitoring in order to produce goal-fulfilling plans that maximize the net benefit accrued.

As the number of applications for human-robot teaming continue to rise, there is an increasing need for planning technologies that can guide robots in such teaming scenarios. In this article, we focus on adapting planning technology to Urban Search And Rescue (USAR) with a human-robot team. We start by showing that several aspects of state-of-the-art planning technology, including temporal planning, partial satisfaction planning, and replanning, can be gainfully adapted to this scenario. We then note that human-robot teaming also throws up an additional critical challenge, namely, enabling existing planners, which work under closed-world assumptions, to cope with the open worlds that are characteristic of teaming problems such as USAR. In response, we discuss the notion of conditional goals, and describe how we represent and handle a specific class of them called open world quantified goals. Finally, we describe how the planner, and its open world extensions, are integrated into a robot control architecture, and provide an empirical evaluation over USAR experimental runs to establish the effectiveness of the planning components.

Autonomous robots that operate in real-world domains face multiple challenges that make planning and goal selection difficult. Not only must planning and execution occur in real time, newly acquired knowledge can invalidate previous plans, and goals and their utilities can change during plan execution. However, these events can also provide opportunities, if the architecture is designed to react appropriately. We present here an architecture that integrates the SapaReplan planner with the DIARC robot architecture, allowing the architecture to react dynamically to changes in the robot's goal structures.