European and American experts teach the robot various movements A group of engineers from Switzerland, Germany and the USA, trained the robot ANYmal walk, run and recover from falls using artificial intelligence algorithms, not manual algorithms-controllers. A feature of the work also lies in the fact that the algorithm was trained during the simulation, and then learned the skill were transferred to the real robot, say the authors of the article in Science is Robotics. Engineers have already created many of the walking robots of different designs, including some that are quite sophisticated. For example, a well known robotics company Boston Dynamics, is able to stay on his feet after the attacks. However, in addition to hardware component is not less important software.

The Holy Grail for data scientists is the ability to obtain labeled data sets for the purpose of training a supervised machine learning algorithm. An algorithm's ability to "learn" is based on training it using a labeled training set -- having known response variable values that correspond to a number of predictor variable values. There are a number of common and maybe not-so-common methods for labeling a data set. In this article, we'll run down a short list of such methods and then you can choose the best for your specific circumstances. Sometimes, labeled datasets are readily available as a byproduct of on-going business operations.

Tokyo AI company Exawizards is using deep learning to analyze unstructured nursing care data such as audio and video recordings. Faced with a rapidly aging population, Japan is turning to new solutions for a many-faceted problem. Amid a low birthrate, more than one-fifth of Japanese are now 70 or older, according to government data. Meanwhile, there are not enough people to care for this cohort. By 2025, there will be a shortfall of 340,000 nursing care workers. Under the Abe administration, Japan has actively welcomed foreign professionals as one measure to address this issue.

Two out of three Indians count on agriculture as their primary livelihood yet the sector contributes just one-sixth of the country's national income. For long, policy mandarins and economists have bemoaned this skew and the urgent need to boost farm productivity but little has moved the needle in Indian farming in recent decades except in pockets. Like in every other sector, artificial intelligence and machine learning techniques, combined with on the ground automated sensing using internet of things devices, is being deployed in agriculture, too. The start-ups are paving the way for tech to to help the Indian farmer to tackle one the biggest challenges before farming: uncertainty. "Uncertainty is the poison in the blood of Indian farming. Farming is difficult and stressful, driving farmers out of farming and sometimes even to suicide. Technology companies in the agri-tech space are helping to make farming into a more stable and desirable industry," says Kahn, a Harvard MBA with over a decade in the Indian agriculture space Today a handful of startups working on AI-backed solutions are paving the way ahead for bringing in the tech to help the Indian farmer to tackle one of the biggest challenges: uncertainty.

Artificial intelligence (AI) has been the talk of the start-up town for a while now. E-commerce major Flipkart plans to create an AI unit called AIForIndia. Online transportation network company Ola recently rolled out an AI-based real-time tracking mechanism for passenger safety. But Unicorns and established players aren't the only ones bullish on this technology. There is a clutch of newbie ventures that have focused their business models primarily around AI. From leveraging artificial intelligence for those in search of a life partner to deploying it for curating gourmet and healthy food menus, start-ups are utilising the much-in-demand technology in the most unique ways.

Previous work on predicting or generating 3D human pose sequences regresses either joint rotations or joint positions. The former strategy is prone to error accumulation along the kinematic chain, as well as discontinuities when using Euler angles or exponential maps as parameterizations. The latter requires re-projection onto skeleton constraints to avoid bone stretching and invalid configurations. This work addresses both limitations. QuaterNet represents rotations with quaternions and our loss function performs forward kinematics on a skeleton to penalize absolute position errors instead of angle errors. We investigate both recurrent and convolutional architectures and evaluate on short-term prediction and long-term generation. For the latter, our approach is qualitatively judged as realistic as recent neural strategies from the graphics literature. Our experiments compare quaternions to Euler angles as well as exponential maps and show that only a very short context is required to make reliable future predictions. Finally, we show that the standard evaluation protocol for Human3.6M produces high variance results and we propose a simple solution.

This paper presents an unsupervised method to learn a neural network, namely an explainer, to interpret a pre-trained convolutional neural network (CNN), i.e., the explainer uses interpretable visual concepts to explain features in middle conv-layers of a CNN. Given feature maps of a conv-layer of the CNN, the explainer performs like an auto-encoder, which decomposes the feature maps into object-part features. The object-part features are learned to reconstruct CNN features without much loss of information. We can consider the disentangled representations of object parts a paraphrase of CNN features, which help people understand the knowledge encoded by the CNN. More crucially, we learn the explainer via knowledge distillation without using any annotations of object parts or textures for supervision. In experiments, our method was widely used to interpret features of different benchmark CNNs, and explainers significantly boosted the feature interpretability without hurting the discrimination power of the CNNs.

Deep neural networks (DNNs) have achieved superior performance in various prediction tasks, but can be very vulnerable to adversarial examples or perturbations. Therefore, it is crucial to measure the sensitivity of DNNs to various forms of perturbations in real applications. We introduce a novel perturbation manifold and its associated influence measure to quantify the effects of various perturbations on DNN classifiers. Such perturbations include various external and internal perturbations to input samples and network parameters. The proposed measure is motivated by information geometry and provides desirable invariance properties. We demonstrate that our influence measure is useful for four model building tasks: detecting potential 'outliers', analyzing the sensitivity of model architectures, comparing network sensitivity between training and test sets, and locating vulnerable areas. Experiments show reasonably good performance of the proposed measure for the popular DNN models ResNet50 and DenseNet121 on CIFAR10 and MNIST datasets.

Recently, deep learning based natural language processing techniques are being extensively used to deal with spam mail, censorship evaluation in social networks, among others. However, there is only a couple of works evaluating the vulnerabilities of such deep neural networks. Here, we go beyond attacks to investigate, for the first time, universal rules, i.e., rules that are sample agnostic and therefore could turn any text sample in an adversarial one. In fact, the universal rules do not use any information from the method itself (no information from the method, gradient information or training dataset information is used), making them black-box universal attacks. In other words, the universal rules are sample and method agnostic. By proposing a coevolutionary optimization algorithm we show that it is possible to create universal rules that can automatically craft imperceptible adversarial samples (only less than five perturbations which are close to misspelling are inserted in the text sample). A comparison with a random search algorithm further justifies the strength of the method. Thus, universal rules for fooling networks are here shown to exist. Hopefully, the results from this work will impact the development of yet more sample and model agnostic attacks as well as their defenses, culminating in perhaps a new age for artificial intelligence.

Consider the problem: given data pair $(\mathbf{x}, \mathbf{y})$ drawn from a population with $f_*(x) = \mathbf{E}[\mathbf{y} | \mathbf{x} = x]$, specify a neural network and run gradient flow on the weights over time until reaching any stationarity. How does $f_t$, the function computed by the neural network at time $t$, relate to $f_*$, in terms of approximation and representation? What are the provable benefits of the adaptive representation by neural networks compared to the pre-specified fixed basis representation in the classical nonparametric literature? We answer the above questions via a dynamic reproducing kernel Hilbert space (RKHS) approach indexed by the training process of neural networks. We show that when reaching any local stationarity, gradient flow learns an adaptive RKHS representation, and performs the global least squares projection onto the adaptive RKHS, simultaneously. In addition, we prove that as the RKHS is data-adaptive and task-specific, the residual for $f_*$ lies in a subspace that is smaller than the orthogonal complement of the RKHS, formalizing the representation and approximation benefits of neural networks.