Chapter 12. Learning Reasoning Strategies in End-to-End Differentiable Proving

Attempts to render deep learning models interpretable, data-efficient, and robust have seen some success through hybridisation with rule-based systems, for example, in Neural Theorem Provers (NTPs). These neuro-symbolic models can induce interpretable rules and learn representations from data via back-propagation, while providing logical explanations for their predictions. However, they are restricted by their computational complexity, as they need to consider all possible proof paths for explaining a goal, thus rendering them unfit for large-scale applications. We present Conditional Theorem Provers (CTPs), an extension to NTPs that learns an optimal rule selection strategy via gradient-based optimisation. We show that CTPs are scalable and yield state-of-the-art results on the CLUTRR dataset, which tests systematic generalisation of neural models by learning to reason over smaller graphs and evaluating on larger ones. Finally, CTPs show better link prediction results on standard benchmarks in comparison with other neural-symbolic models, while being explainable. All source code and datasets are available online. (At https://github.com/uclnlp/ctp)

A Piecewise Yield Failure Criterion including the Critical State for Brittle Rock

The critical state of rock is an important index for measuring the changes in rock characteristics. However, this state is not unique because of the different researcher assumptions. Based on the theory of the partial differential equation proposed by Vutukuri, according to Mohr’s envelope, a piecewise yield failure criterion (referred to as the Mohr–Wedge criterion), including the critical state for brittle rock, is obtained by introducing the wedge model to solve this equation. The Mohr–Wedge (M–W) criterion consisting of nonlinear and linear components includes the critical state for brittle rock. When the minimum principal stress σ3 is lower than the confining pressure σk, the maximum principal stress σ1 varies nonlinearly with σ3; otherwise, σ1 varies linearly with σ3. This variation conforms to rock deformation features under triaxial compression. In this study, we investigate the rationality of this critical state by an analogy method and illustrate that the critical state mentioned in this criterion is related to the microcracks in the potential failure zone of the rock. Alternatively, the primary object of this study is to reveal the applicability of predicting the yield state for this criterion. The method used in our study is compared to the Mohr–Coulomb (M-C) criterion, the Hoek–Brown (H-B) criterion, and the Exponential (Exp.) criterion by the yield surfaces on the deviatoric plane. Notably, there is a vertex consistent region for the four criteria, but except for this region, the yield state of rock predicted by the four criteria is quite different, depending on the extent of the parameters for the criteria and the magnitude of the slopes of the yield surfaces. The results show that the M-W criterion has certain applicability for predicting the rock yield state by using the multiple data of rock triaxial compression tests in the published literature.

EHSOD: CAM-Guided End-to-End Hybrid-Supervised Object Detection with Cascade Refinement

Object detectors trained on fully-annotated data currently yield state of the art performance but require expensive manual annotations. On the other hand, weakly-supervised detectors have much lower performance and cannot be used reliably in a realistic setting. In this paper, we study the hybrid-supervised object detection problem, aiming to train a high quality detector with only a limited amount of fully-annotated data and fully exploiting cheap data with image-level labels. State of the art methods typically propose an iterative approach, alternating between generating pseudo-labels and updating a detector. This paradigm requires careful manual hyper-parameter tuning for mining good pseudo labels at each round and is quite time-consuming. To address these issues, we present EHSOD, an end-to-end hybrid-supervised object detection system which can be trained in one shot on both fully and weakly-annotated data. Specifically, based on a two-stage detector, we proposed two modules to fully utilize the information from both kinds of labels: 1) CAM-RPN module aims at finding foreground proposals guided by a class activation heat-map; 2) hybrid-supervised cascade module further refines the bounding-box position and classification with the help of an auxiliary head compatible with image-level data. Extensive experiments demonstrate the effectiveness of the proposed method and it achieves comparable results on multiple object detection benchmarks with only 30% fully-annotated data, e.g. 37.5% mAP on COCO. We will release the code and the trained models.

Suicide Risk Assessment via Temporal Psycholinguistic Modeling (Student Abstract)

Social media platforms are increasingly being used for studying psycho-linguistic phenomenon to model expressions of suicidal intent in tweets. Most recent work in suicidal ideation detection doesn't leverage contextual psychological cues. In this work, we hypothesize that the contextual information embedded in the form of historical activities of users and homophily networks formed between like-minded individuals in Twitter can substantially improve existing techniques for automated identification of suicidal tweets. This premise is extensively tested to yield state of the art results as compared to linguistic only models, and the state-of-the-art model.

Leveraging Multi-View Image Sets for Unsupervised Intrinsic Image Decomposition and Highlight Separation

We present an unsupervised approach for factorizing object appearance into highlight, shading, and albedo layers, trained by multi-view real images. To do so, we construct a multi-view dataset by collecting numerous customer product photos online, which exhibit large illumination variations that make them suitable for training of reflectance separation and can facilitate object-level decomposition. The main contribution of our approach is a proposed image representation based on local color distributions that allows training to be insensitive to the local misalignments of multi-view images. In addition, we present a new guidance cue for unsupervised training that exploits synergy between highlight separation and intrinsic image decomposition. Over a broad range of objects, our technique is shown to yield state-of-the-art results for both of these tasks.

GeniePath: Graph Neural Networks with Adaptive Receptive Paths

We present, GeniePath, a scalable approach for learning adaptive receptive fields of neural networks defined on permutation invariant graph data. In GeniePath, we propose an adaptive path layer consists of two complementary functions designed for breadth and depth exploration respectively, where the former learns the importance of different sized neighborhoods, while the latter extracts and filters signals aggregated from neighbors of different hops away. Our method works in both transductive and inductive settings, and extensive experiments compared with competitive methods show that our approaches yield state-of-the-art results on large graphs.

Crop Recommendation System to Maximize Crop Yield in Ramtek region using Machine Learning

In Indian economy and employment agriculture plays major role. The most common problem faced by the Indian farmers is they do not opt crop based on the necessity of soil, as a result they face serious setback in productivity. This problem can be addressed through precision agriculture. This method takes three parameters into consideration, viz: soil characteristics, soil types and crop yield data collection based on these parameters suggesting the farmer suitable crop to be cultivated. Precision agriculture helps in reduction of non suitable crop which indeed increases productivity, apart from the following advantages like efficacy in input as well as output and better decision making for farming. This method gives solutions like proposing a recommendation system through an ensemble model with majority voting techniques using random tree, CHAID, K _ Nearest Neighbour and Naive Bayes as learner to recommend suitable crop based on soil parameters with high specific accuracy and efficiency. The classified image generated by these techniques consists of ground truth statistical data and parameters of it are weather, crop yield, state and district wise crops to predict the yield of a particular crop under particular weather condition.

A Screening Method for Prevention of Ratcheting Strain Derived From Movement of Temperature Distribution

In this paper, we simplify the existing method and propose a screening method to prevent thermal ratcheting strain in the design of practical components. The proposed method consists of two steps to prevent the continuous accumulation of ratcheting strain. The first step is to determine whether all points through the wall thickness are in the plastic state. This is based on an equivalent membrane stress, which comprises the primary stress and the secondary membrane stress. When the equivalent stress exceeds the yield strength in some regions of the cylinder, the axial lengths of these regions are measured for the second step. The second step is to determine whether the accumulation of the plastic strain saturates. For this purpose, we define the screening criteria for the length of the area with full section yield state. When this length is sufficiently small, residual stress is generated in the direction opposite to the plastic deformation direction. As a result of residual stress, further accumulation of the plastic deformation is suppressed, and finally shakedown occurs. To validate the proposed method, we performed a set of elastoplastic finite element method (FEM) analyses, with the assumption of elastic perfectly plastic material. Not only did we investigate about the effect of the axial length of the area with full section yield state but also we investigated about effects of spatial distribution of temperature, existence of primary stress, and radius thickness ratio.

The Evolution of Stress and Strain around the Bayan Har Block in the Tibetan Plateau

With the constraint of GPS observation, the tectonic deformation of the Bayan Har block and its periphery faults is investigated based on an elastoplastic plane-stress finite element model. The results show that the elastic model cannot explain the current GPS observation in the Bayan Har block. When East Kunlun fault and Yushu-Xianshuihe fault are under plastic yield state or high strain localization, the calculated velocities fit well with the observation values. It indicates that most of the current shear deformations or strain localizations are absorbed by these two large strike-slip faults. In addition, if the recurrence intervals of large earthquakes are used to limit the relative yield strength of major faults, the order of entering the plastic yield state of the major faults around Bayan Har block is as follows. The first faults to enter the yield state are Yushu-Xianshuihe faults and the middle segment of East Kunlun faults. Then, Margaichaka-RolaKangri faults (Mani segment) and Heishibeihu faults would enter the yield state. The last faults to enter the yield state are the eastern segment of East Kunlun faults and Longmenshan faults, respectively. These results help us to understand the slip properties of faults around the southeastward moving Bayan Har block.