15 research outputs found
NL to PDDL: One-Shot Learning of Planning Domains from Natural Language Process Manuals
No full textAutomated Planning (AP) is a key component of Artificial General Intelligence and has been successfully employed in applications ranging from scheduling observations of Hubble Space Telescope to generating dialogue agents. A significant bottleneck for its widespread adoption is acquiring accurate domain models which formally encode the planning problem’s environment. Traditionally, these domain models have been hand-coded by human experts and knowledge engineers. However, manually encoding domain models is an increasingly difficult task when one moves away from toy domains towards complex real-world problem scenarios.To resolve this, the AP community has developed several systems to automatically acquire domain models from valid sequences of actions called plans. This approach has two significant issues. First, the generated domain models might be incomplete, error-prone, and hard to understand and/or modify. Second, most domain learning approaches are based on data-intensive inductive learning, which needs large quantities of structured data (plans) to converge. This data is seldom available without an accurate domain model, which leads to a causality dilemma.To mitigate these issues, we take advantage of readily available and easy to craft Natural Language (NL) data. We present a pipeline called NLtoPDDL, which takes as input a classical domain’s process manual written in a natural language and outputs its Planning Domain Definition Language (PDDL) model. Specifically, NLtoPDDL does this in two steps: first, it combines pre-trained contextual embeddings with an approach developed in previous research, called EASDRL that extracted structured plans from NL data using Deep Reinforcement Learning (DRL), and a consequence, NLtoPDDL beats the EASDRL model which is the current state-of-the-art on action sequence extraction problem; second, it uses the trained DRL model from the first step to extract structured plans from a domain process manual and employs a modified Learning Object Centered Models (LOCM2) algorithm to one-shot learn a PDDL model. Finally, we showcase the effectiveness of our pipeline on four planning domains of varying complexities, by evaluating our learned domain models for soundness, completeness, validity and intuitiveness
NLtoPDDL: One-Shot Learning of PDDL Models from Natural Language Process Manuals
No full textExisting automated domain acquisition approaches require large amounts of structured data in the form of plans or plan traces to converge. Further, automatically-generated domain models can be incomplete, error-prone, and hard to understand or modify. To mitigate these issues, we take advantage of readily-available natural language data: existing process manuals. We present a domain-authoring pipeline called NLtoPDDL, which takes as input a plan written in natural language and outputs a corresponding PDDL model. We employ a two-stage approach: stage one advances the state-of-the-art in action sequence extraction by utilizing transfer learning via pre-trained contextual language models (BERT and ELMo). Stage two employs an interactive modification of an object-centric algorithm which keeps human-in-the-loop to one-shot learn a PDDL model from the extracted plan. We show that NLtoPDDL is an effective and flexible domain-authoring tool by using it to learn five real-world planning domains of varying complexities and evaluating them for their completeness, soundness and quality.Algorithmic
Economic impact of micro irrigation adoption scheme "Per Drop More Crop" (PDMC): A case of sugarcane, banana and cotton cultivation in Maharashtra (India)
No full textThis is a working draft. Please do not cite without permission of the author
Methane Leak Quantification on Edge Devices Using Deep Learning
No full textOil and gas domain deals with a varied set of problems ranging from methane leaks to prognostics and health management. This study demonstrates a solution to identify whether there is a methane leak or not and classify on what level the leak occurs based on different flow rates (5.3 g/h – 2051.6 g/h) at 5 distances (4.6 m - 15.6 m), using various spatial and temporal preprocessing techniques and deep learning models. For this study, we are using the GasVid methane leak dataset which consists of videos taken from infrared cameras on 2 different separators with a frame rate of 15 frames per second. Firstly, we applied a series of preprocessing steps, including contrast enhancement, Gaussian blur, three different background subtraction methods, namely moving average background subtraction method, KNN Gaussian mixture model, and Gunnar Farneback optical flow analysis. To save computational resources, temporal down-sampling was applied on the video frames. Thereafter, experiments were conducted using the 3D CNN model by modifying hyperparameters. It was found that on comparing the Adam and Lion (EvoLved Sign Momentum) optimizers, the Lion Optimizer increased accuracy by more than 34% and achieved state-of-the-art accuracy of 41.36% on 4.6 m videos. Further, the moving average background subtraction method\u27s performance surpassed other background subtraction techniques. In addition, applying spatial preprocessing and down-sampling raw videos were compressed from 2.3GB to ~200MB which is a reduction factor of 11
STAR: Superhuman training in augmented reality
No full textThe goals of Superhuman Sports games include to involve physical activity and enhance the skills and abilities of a human through technology. They are played for fun, competition or to improve the players’ health condition. To meet these goals, we designed and developed STAR: Superhuman Training in Augmented Reality, an augmented reality adventure shooter, and implemented it on the Microsoft Hololens. Our game promotes physical activity by making you avoid dangerous enemies and gather energy to deal with this threat while navigating a narrow path above lava. Social interaction is stimulated by its multiplayer mode, in which players have to work together to destroy an energy core. Player testing showed that we achieve our goal of physical exercise by making the player move at a pace slightly less than brisk walking and that the game is fun and immersive. These results show that STAR is a promising step in the right direction for the development of superhuman sports using augmented reality.Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Computer Graphics and VisualisationSystem Engineerin
Soil-Structure Interaction and Aseismic Design of a Stadium Building
No full textThe paper presents case history of a reinforced concrete stadium building which had been structurally designed for a particular component configuration and also constructed upto seating level and which was referred to the author for suggesting structural modifications and redesigning for different configuration which meant curtailing middle two main columns each above the seating level out of four columns in each of left and right halves of the building. The required modifications necessitated analysis of the modified frame under static loads taking into account soil-structure interaction. The other problem to be tackled was ensuring lateral stability with reduced number of main columns, which are slender and have restriction in size, under earthquake conditions. Since the structure could have free vibrations in coupled translation and yawing, advantage has been taken of stiffness of rear columns whose size was not restricted
Comparing learned representations of deep neural networks
No full textThis electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019Cataloged from student-submitted PDF version of thesis.Includes bibliographical references (pages 63-64).In recent years, a variety of deep neural network architectures have obtained substantial accuracy improvements in tasks such as image classification, speech recognition, and machine translation, yet little is known about how different neural networks learn. To further understand this, we interpret the function of a deep neural network used for classification as converting inputs to a hidden representation in a high dimensional space and applying a linear classifier in this space. This work focuses on comparing these representations as well as the learned input features for different state-of-the-art convolutional neural network architectures. By focusing on the geometry of this representation, we find that different network architectures trained on the same task have hidden representations which are related by linear transformations. We find that retraining the same network architecture with a different initialization does not necessarily lead to more similar representation geometry for most architectures, but the ResNeXt architecture consistently learns similar features and hidden representation geometry. We also study connections to adversarial examples and observe that networks with more similar hidden representation geometries also exhibit higher rates of adversarial example transferability.by Vivek N. Miglani.M. Eng.M.Eng. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Scienc
Legality of Targeting Satellites under Jus in Bello: Specific Focus on Non-kinetic ASAT weapons
No full textPurpose – Satellites are attractive military objectives due to their trajectorial predictability and essential functions they provide to military operations. In the last 13 years, at least three States (namely, USA, China and India) have successfully conducted kinetic anti-satellite (ASAT) missile tests which significantly increased amount of low-Earth orbit space debris, some of which are still orbiting and pose threat to space assets (Miglani, 2019, Wolf, 2007). All of these ASAT weapon tests were conducted against the self-owned space assets of the state conducting the test, therefore, these events did not trigger application of the law of armed conflict (jus in bello). However, that does not mean that legal evaluation of these tests, especially in terms of jus in bello, is practically insignificant, bearing in mind that technical destructive capabilities are already present and legitimacy of the use of these weapons is not evident. Indeed, some authors have already stressed out difficulties of legitimizing kinetic ASAT weapons, or, to be more precise, armed attacks against space assets. It has been argued that kinetic ASAT attacks in some cases could hardly fit principle of proportionality due to unpredictability of the amount of space debris and secondary collateral damage a blast-generated space debris could potentially cause (Stephens and Steer, 2016) or even attacks themselves in some cases might be indiscriminate in nature (Koplow, 2009). It could be observed that legitimacy of ASAT weapons is questionable mainly due to effects of kinetic attacks, but there are weapons, which aim to jam communication systems or cause malfunction with directed energy without generating space debris, except probably one inactive orbiting satellite. Therefore, most of the arguments applicable to kinetic ASAT attacks may not be applied to non-kinetic ones. In this article, the author argues that the use of non-kinetic ASAT weapons in certain conditions is hardly compatible with general principles of jus in bello, especially rules of targeting. The purpose of this article is to analyze whether the use of non-kinetic anti-satellite weapons during armed conflict is in accordance with jus in bello and, if not, what are conditions of their legitimate use. Design/methodology/approach – this piece of research is based on information analysis, linguistic, systemic analysis and analogy methods. Research covering aspects of the law of outer space warfare will be analyzed and systemized while linguistic method is a helpful tool to interpret statutory rules governing weaponization. Analogy method is used to disclose definition of non-kinetic ASAT attacks using arguments applied to cyberattacks. Finding – the use of non-kinetic ASAT weapons has limits under jus in bello. Research limitations/implications – research is limited to non-kinetic weapons. This article does not disclose detailed technical aspects of non-kinetic ASAT weapons. It only highlights capabilities and function of these weapons and legal implications that the use those weapons against objects in outer space. Protection of persons under jus in bello, including targeting rules related with humans as targets, is not the object of this article. Practical implications – since kinetic and non-kinetic space weapons are already present and still being developed, this research could contribute to determine legal boundaries of satellite attacks in practice. Originality/Value – the focus on non-kinetic ASAT weapons is novel, since most of the research involves legal issues related to the effects of the use of kinetic ASAT weapons
Public understanding of food risks in four European countries: a qualitative study
No full textBackground: In the wake of the ` bovine spongiform encephalopathy ( BSE) crisis' there was renewed interest in how those responsible for public health could take account of public views, both to ` democratize' policy making and to increase the likelihood of information about health risks resonating with public concerns. This study explored how members of the public in four European countries ( Finland, Germany, Italy and the UK) understood food risks in general, and risks arising from BSE in particular. The aims were to identify the sources of knowledge used and trusted by the public and to explore how public views could be accessed for public health information policy.
Methods: Thirty- six focus group interviews were held using a common protocol across the four countries, including people from four lifecycle stages.
Results: The study demonstrated the utility of using focus groups as a relatively efficient method for accessing public views, and the feasibility of cross- national qualitative research on public views. We found that public views of food risks are neither irrational nor naive, but that they do need to be interpreted in the context of everyday food purchasing decisions, in which particular food risks are unlikely to have the same salience as they do for experts.
Conclusions: Focus groups are a feasible method for accessing public knowledge on public health risks to inform information strategies
