This paper investigates the utilization of a physics simulation environment as the imagination of a robot, where it creates a replica of the detected terrain in a physics simulation environment in its memory, and "imagines" a simulated version of itself in that memory, performing actions and navigation on the terrain. The physics of the environment simulates the movement of robot parts and its interaction with the objects in the environment and the terrain, thus avoiding the need for explicitly programming many calculations.

Neural module networks (NMNs) are a popular approach for modeling compositionality: they achieve high accuracy when applied to problems in language and vision, while reflecting the compositional structure of the problem in the network architecture. However, prior work implicitly assumed that the structure of the network modules, describing the abstract reasoning process, provides a faithful explanation of the model's reasoning; that is, that all modules perform their intended behaviour. In this work, we propose and conduct a systematic evaluation of the intermediate outputs of NMNs on NLVR2 and DROP, two datasets which require composing multiple reasoning steps. We find that the intermediate outputs differ from the expected output, illustrating that the network structure does not provide a faithful explanation of model behaviour. To remedy that, we train the model with auxiliary supervision and propose particular choices for module architecture that yield much better faithfulness, at a minimal cost to accuracy.

This chapter describes how the SP System, meaning the SP Theory of Intelligence, and its realisation as the SP Computer Model, may promote transparency and granularity in AI, and some other areas of application. The chapter describes how transparency in the workings and output of the SP Computer Model may be achieved via three routes: 1) the program provides a very full audit trail for such processes as recognition, reasoning, analysis of language, and so on. There is also an explicit audit trail for the unsupervised learning of new knowledge; 2) knowledge from the system is likely to be granular and easy for people to understand; and 3) there are seven principles for the organisation of knowledge which are central in the workings of the SP System and also very familiar to people (eg chunking-with-codes, part-whole hierarchies, and class-inclusion hierarchies), and that kind of familiarity in the way knowledge is structured by the system, is likely to be important in the interpretability, explainability, and transparency of that knowledge. Examples from the SP Computer Model are shown throughout the chapter.

Combinatorial optimization plays an important role in real-world problem solving. In the big data era, the dimensionality of a combinatorial optimization problem is usually very large, which poses a significant challenge to existing solution methods. In this paper, we examine the generalization capability of a machine learning model for problem reduction on the classic travelling salesman problems (TSP). We demonstrate that our method can greedily remove decision variables from an optimization problem that are predicted not to be part of an optimal solution. More specifically, we investigate our model's capability to generalize on test instances that have not been seen during the training phase. We consider three scenarios where training and test instances are different in terms of: 1) problem characteristics; 2) problem sizes; and 3) problem types. Our experiments show that this machine learning based technique can generalize reasonably well over a wide range of TSP test instances with different characteristics or sizes. While the accuracy of predicting unused variables naturally deteriorates as a test instance is further away from the training set, we observe that even when tested on a different TSP problem variant, the machine learning model still makes useful predictions about which variables can be eliminated without significantly impacting solution quality.

Challenging problems such as open-domain question answering, fact checking, slot filling and entity linking require access to large, external knowledge sources. While some models do well on individual tasks, developing general models is difficult as each task might require computationally expensive indexing of custom knowledge sources, in addition to dedicated infrastructure. To catalyze research on models that condition on specific information in large textual resources, we present a benchmark for knowledge-intensive language tasks (KILT). All tasks in KILT are grounded in the same snapshot of Wikipedia, reducing engineering turnaround through the re-use of components, as well as accelerating research into task-agnostic memory architectures. We test both task-specific and general baselines, evaluating downstream performance in addition to the ability of the models to provide provenance. We find that a shared dense vector index coupled with a seq2seq model is a strong baseline, outperforming more tailor-made approaches for fact checking, open-domain question answering and dialogue, and yielding competitive results on entity linking and slot filling, by generating disambiguated text. KILT data and code are available at https://github.com/facebookresearch/KILT.

MR imaging is a powerful and diverse imaging technique employed to investigate and diagnose a range of diseases in different body areas. MRI scans are acquired by employing specific parameters in a "sequence" to encode in data in arbitrary space known as "k-space". Image is reconstructed by applying mathematical transforms (mainly Fourier) to the k-space data. To obtain images of particular contrast (T1w, T2w, T2* etc) optimal sequence settings must be employed. Briefly, image contrast arise from magnetic property of the hydrogen atoms, that can varied setting such as echo time and Tr in sequence setting.

This paper describes problems in AI research and how the SP System may help to solve them. Most of the problems are described by leading researchers in AI in interviews with science writer Martin Ford, and reported by him in his book Architects of Intelligence. These problems, each with potential solutions via SP, are: how to overcome the divide between symbolic and non-symbolic kinds of knowledge and processing; eliminating large and unexpected errors in recognition; the challenge of unsupervised learning; the problem of generalisation, with under- and over-generalisation; learning from a single exposure or experience; the problem of transfer learning; how to create learning that is fast, economical in demands for data and computer resources; the problems of transparency in results and processing; problems in the processing of natural language; problems in the development of probabilistic reasoning; the problem of catastrophic forgetting; how to achieve generality across several aspects of AI. The SP System provides a relatively promising foundation for the development of artificial general intelligence.

As a field of AI, Machine Reasoning (MR) uses largely symbolic means to formalize and emulate abstract reasoning. Studies in early MR have notably started inquiries into Explainable AI (XAI) -- arguably one of the biggest concerns today for the AI community. Work on explainable MR as well as on MR approaches to explainability in other areas of AI has continued ever since. It is especially potent in modern MR branches, such as argumentation, constraint and logic programming, planning. We hereby aim to provide a selective overview of MR explainability techniques and studies in hopes that insights from this long track of research will complement well the current XAI landscape. This document reports our work in-progress on MR explainability.

A common approach to solve complex tasks is by breaking them down into simple sub-problems that can then be solved by simpler modules. However, these approaches often need to be designed and trained specifically for each complex task. We propose a general approach, Text Modular Networks(TMNs), where the system learns to decompose any complex task into the language of existing models. Specifically, we focus on Question Answering (QA) and learn to decompose complex questions into sub-questions answerable by existing QA models. TMNs treat these models as blackboxes and learn their textual input-output behavior (i.e., their language) through their task datasets. Our next-question generator then learns to sequentially produce sub-questions that help answer a given complex question. These sub-questions are posed to different existing QA models and, together with their answers, provide a natural language explanation of the exact reasoning used by the model. We present the first system, incorporating a neural factoid QA model and a symbolic calculator, that uses decomposition for the DROP dataset, while also generalizing to the multi-hop HotpotQA dataset. Our system, ModularQA, outperforms a cross-task baseline by 10-60 F1 points and performs comparable to task-specific systems, while also providing an easy-to-read explanation of its reasoning.

Knowledge-based Visual Question Answering (KVQA) requires external knowledge beyond the visible content to answer questions about an image. This ability is challenging but indispensable to achieve general VQA. One limitation of existing KVQA solutions is that they jointly embed all kinds of information without fine-grained selection, which introduces unexpected noises for reasoning the correct answer. How to capture the question-oriented and information-complementary evidence remains a key challenge to solve the problem. Inspired by the human cognition theory, in this paper, we depict an image by multiple knowledge graphs from the visual, semantic and factual views. Thereinto, the visual graph and semantic graph are regarded as image-conditioned instantiation of the factual graph. On top of these new representations, we re-formulate Knowledge-based Visual Question Answering as a recurrent reasoning process for obtaining complementary evidence from multimodal information. To this end, we decompose the model into a series of memory-based reasoning steps, each performed by a G raph-based R ead, U pdate, and C ontrol ( GRUC ) module that conducts parallel reasoning over both visual and semantic information. By stacking the modules multiple times, our model performs transitive reasoning and obtains question-oriented concept representations under the constrain of different modalities. Finally, we perform graph neural networks to infer the global-optimal answer by jointly considering all the concepts. We achieve a new state-of-the-art performance on three popular benchmark datasets, including FVQA, Visual7W-KB and OK-VQA, and demonstrate the effectiveness and interpretability of our model with extensive experiments.