Plants have become an important source of energy, and are a fundamental piece in the puzzle to solve the problem of global warming. However, plant diseases are threatening the livelihood of this important source. Convolutional neural networks (CNN) have demonstrated great performance (beating that of humans) in object recognition and image classification problems. This paper describes the feasibility of CNN for plant disease classification for leaf images taken under the natural environment. The model is designed based on the LeNet architecture to perform the soybean plant disease classification. 12,673 samples containing leaf images of four classes, including the healthy leaf images, were obtained from the PlantVillage database. The images were taken under uncontrolled environment. The implemented model achieves 99.32% classification accuracy which show clearly that CNN can extract important features and classify plant diseases from images taken in the natural environment.

Negation plays a significant role in spoken and written natural languages. Negation is used in language to deny something or to reverse the polarity or the sense of a statement. This paper presents a novel approach to automatically handling negation in tutorial dialogues using deep learning methods. In particular, we explored various Long Short Term Memory (LSTM) models to automatically detect negation focus, scope and cue in tutorial dialogues collected from experiments with actual students interacting with the state-of-the-art intelligent tutoring system, DeepTutor. The results obtained are promising.

Pollen recognition has a wide range of industrial and scientific applications. It guides the energy industry to potential oil and gas deposits, it is proxy data for climate-change scien- tists, and it increases agricultural production. However, pollen recognition is time consuming because it is usually done by visual inspection. Current automated solutions rely on pre-designed measurements of texture and contours, which require tuning for optimal features of a dataset. Also, most methods classify pollen using single-focus images, which require pollen grains to be captured at specific focal planes. We take a difference approach. Instead of using single-focus images, we use stacks of multifocal images (i.e., z-stack) to account for both visual characteristics and 3-D information. We automatically learn from the data the best visual characteristics for classifying pollen using deep-learning methods. Here, we train convolutional and recurrent neural networks (CNN and RNN) to learn the optimal features and recognize a pollen grain as a sequence of multifocal images acquired by an optical microscope. Additionally, we transfer the knowledge pre-trained network to ours to improve its classification and convergence speed. We evaluated our method using 392 stack sequences of 10 types of pollen grains with 10 images for each sequence. Our method achieved a remarkable classi- fication rate of 100%.

In this paper we explore acceleration techniques for large scale nonconvex optimization problems with special focuses on deep neural networks. The extrapolation scheme is a classical approach for accelerating stochastic gradient descent for convex optimization, but it does not work well for nonconvex optimization typically. Alternatively, we propose an interpolation scheme to accelerate nonconvex optimization and call the method Interpolatron. We explain motivation behind Interpolatron and conduct a thorough empirical analysis. Empirical results on DNNs of great depths (e.g., 98-layer ResNet and 200-layer ResNet) on CIFAR-10 and ImageNet show that Interpolatron can converge much faster than the state-of-the-art methods such as the SGD with momentum and Adam. Furthermore, Anderson's acceleration, in which mixing coefficients are computed by least-squares estimation, can also be used to improve the performance. Both Interpolatron and Anderson's acceleration are easy to implement and tune. We also show that Interpolatron has linear convergence rate under certain regularity assumptions.

Many problems at the intersection of combinatorics and computer science require solving for a permutation that optimally matches, ranks, or sorts some data. These problems usually have a task-specific, often non-differentiable objective function that data-driven algorithms can use as a learning signal. In this paper, we propose the Sinkhorn Policy Gradient (SPG) algorithm for learning policies on permutation matrices. The actor-critic neural network architecture we introduce for SPG uniquely decouples representation learning of the state space from the highly-structured action space of permutations with a temperature-controlled Sinkhorn layer. The Sinkhorn layer produces continuous relaxations of permutation matrices so that the actor-critic architecture can be trained end-to-end. Our empirical results show that agents trained with SPG can perform competitively on sorting, the Euclidean TSP, and matching tasks. We also observe that SPG is significantly more data efficient at the matching task than the baseline methods, which indicates that SPG is conducive to learning representations that are useful for reasoning about permutations.

A bstract ： The paper intends to model the stability of power system with a deep learning algorithm to the problem, aiming to delay the removal of the fault. The so - called "fail - delay cut - off" refers to the occurrenc e of N - 1 backup protection action on the backbone network of the system, resulting in longer time for the removal of the fault. In practice, through the analysis and calculation of a large number of online data, we have found that the N - 1 failure system of the main protection action will not be unstable, which is also a guarantee of the operation mode arrangement. In the case of the N - 1 backup protection action, there is an approximately 2.5% probability that the system will be destabilized. Therefore, rese arch is needed to improve the operating arrangement.

Recently, it has been shown that deep neural networks (DNN) are subject to attacks through adversarial samples. Adversarial samples are often crafted through adversarial perturbation, i.e., manipulating the original sample with minor modifications so that the DNN model labels the sample incorrectly. Given that it is almost impossible to train perfect DNN, adversarial samples are shown to be easy to generate. As DNN are increasingly used in safety-critical systems like autonomous cars, it is crucial to develop techniques for defending such attacks. Existing defense mechanisms which aim to make adversarial perturbation challenging have been shown to be ineffective. In this work, we propose an alternative approach. We first observe that adversarial samples are much more sensitive to perturbations than normal samples. That is, if we impose random perturbations on a normal and an adversarial sample respectively, there is a significant difference between the ratio of label change due to the perturbations. Observing this, we design a statistical adversary detection algorithm called nMutant (inspired by mutation testing from software engineering community). Our experiments show that nMutant effectively detects most of the adversarial samples generated by recently proposed attacking methods. Furthermore, we provide an error bound with certain statistical significance along with the detection.

Text-based games are suitable test-beds for designing agents that can learn by interaction with the environment in the form of natural language text. Very recently, deep reinforcement learning based agents have been successfully applied for playing text-based games. In this paper, we explore the possibility of designing a single agent to play several text-based games and of expanding the agent's vocabulary using the vocabulary of agents trained for multiple games. To this extent, we explore the application of recently proposed policy distillation method for video games to the text-based game setting. We also use text-based games as a test-bed to analyze and hence understand policy distillation approach in detail.

Enabling robots to autonomously navigate complex environments is essential for real-world deployment. Prior methods approach this problem by having the robot maintain an internal map of the world, and then use a localization and planning method to navigate through the internal map. However, these approaches often include a variety of assumptions, are computationally intensive, and do not learn from failures. In contrast, learning-based methods improve as the robot acts in the environment, but are difficult to deploy in the real-world due to their high sample complexity. To address the need to learn complex policies with few samples, we propose a generalized computation graph that subsumes value-based model-free methods and model-based methods, with specific instantiations interpolating between model-free and model-based. We then instantiate this graph to form a navigation model that learns from raw images and is sample efficient. Our simulated car experiments explore the design decisions of our navigation model, and show our approach outperforms single-step and $N$-step double Q-learning. We also evaluate our approach on a real-world RC car and show it can learn to navigate through a complex indoor environment with a few hours of fully autonomous, self-supervised training. Videos of the experiments and code can be found at github.com/gkahn13/gcg

But the space of all testable correlates is far larger than scientific resources can currently address, so the ability to accurately predict correlations would be useful to plan research and allocate resources. Using a dataset of approximately 170,000 correlational findings extracted from leading social science journals, we show that a trained neural network can accurately predict the reported correlations using only the text descriptions of the correlates. Accurate predictive models such as these can guide scientists towards promising untested correlates, better quantify the information gained from new findings, and has implications for moving artificial intelligence systems from predicting structures to predicting relationships in the real world.