PROPONENTS of education technology have made remarkable promises over the past two decades: that by 2019, half of all secondary school courses would be online; videos and practice problems can let students learn mathematics at their own pace; in 50 years only 10 mega-institutions of higher education would be left; or that typical students left alone with internet-connected computers can learn anything without the help of schools or teachers. Then in 2020, people around the world were forced to turn to online learning as the coronavirus pandemic shut down schools serving more than 1 billion students. It was education technology's big moment, but for many students and families, remote learning has been a disappointment. When the world needs it most, why has education technology seemed so lacklustre? Educational software has a long history, but throughout there have been two major challenges.

This course IS NOT like any other Python Programming course you can take online. At the end of this course you will MASTER all the Python 3 key concepts starting from scratch and you'll be in the top Python Programmers. Welcome to this practical Python Programming course for learning Python, the most in-demand programming languages across the job market in 2019. "This is the only course you need in order to MASTER every key aspect of Python. Don't look for other Python courses." by Daniel A. "This Python course, though I am still half way through, is the best I have seen so far, that is why I am giving it a 5 star. I am enrolled in two more Python courses in Udemy, and this is the most useful. Keep it up!" by Malvin Arceo "This is an excellent course for anyone who wants to learn Python from scratch or just do a refresher of a language. Everything is well explained and lots of quizzes and coding exercises are very helpful. Highly recommended:)" by Tomaso "Overall a great Python course, with lots of extra details added, to make it as comprehensive as possible. At the moment, I consider it the best Python course for anybody who wants to learn more about this subject."

June 8, 2019 Updated: April 20, 2020 "[AI] is going to change the world more than anything in the history of mankind. AI oracle and venture capitalist Dr. Kai-Fu Lee, 2018 In a nondescript building close to downtown Chicago, Marc Gyongyosi and the small but growing crew of IFM / Onetrack.AI have one rule that rules them all: think simple. The words are written in simple font on a simple sheet of paper that's stuck to a rear upstairs wall of their industrial two-story workspace. Sitting at his cluttered desk, located near an oft-used ping-pong table and prototypes of drones from his college days suspended overhead, Gyongyosi punches some keys on a laptop to pull up grainy video footage of a forklift driver operating his vehicle in a warehouse. It was captured from overhead courtesy of a Onetrack.AI "forklift vision system." The Future of Artificial Intelligence Artificial intelligence is impacting the future of virtually every industry and every human being. Artificial intelligence has acted as the main driver of emerging technologies like big data, robotics and IoT, and it will continue to act as a technological innovator for the foreseeable future. Employing machine learning and computer vision for detection and classification of various "safety events," the shoebox-sized device doesn't see all, but it sees plenty. Like which way the driver is looking as he operates the vehicle, how fast he's driving, where he's driving, locations of the people around him and how other forklift operators are maneuvering their vehicles. IFM's software automatically detects safety violations (for example, cell phone use) and notifies warehouse managers so they can take immediate action. The main goals are to prevent accidents and increase efficiency. The mere knowledge that one of IFM's devices is watching, Gyongyosi claims, has had "a huge effect." Marc Gyongyosi Photo Credit: IFM/OneTrack.AI The lower level of IFM was designed to mimic a warehouse environment so products can be effectively tested on site. Photo Credit: IFM/OneTrack.AI "If you think about a camera, it really is the richest sensor available to us today at a very interesting price point," he says. "Because of smartphones, camera and image sensors have become incredibly inexpensive, yet we capture a lot of information.

Multiple choice tests provide test-takers the ability to compare answers to eliminate choices (or guess the correct one). Each choice can be compared with the question to infer patterns that might have been missed; it's arguably the ability to narrow down the right answer from sets of answers that's the test of true comprehension. Inspired by this, researchers at Tel Aviv University and Facebook developed a machine learning model that generates answers to the Raven Progressive Matrix (RPM), a type of intelligence test where the goal is to complete the location in a grid of abstract images. The coauthors claim that their algorithm is not only able to generate a plausible set of answers competitive with state-of-the-art methods, but that it could be used to build an automatic tutoring system that adjusts to the proficiencies of individual students. RPM is a nonverbal test typically used in educational settings like schools.

For a function $\mathsf{K} : \mathbb{R}^{d} \times \mathbb{R}^{d} \to \mathbb{R}_{\geq 0}$, and a set $P = \{ x_1, \ldots, x_n\} \subset \mathbb{R}^d$ of $n$ points, the $\mathsf{K}$ graph $G_P$ of $P$ is the complete graph on $n$ nodes where the weight between nodes $i$ and $j$ is given by $\mathsf{K}(x_i, x_j)$. In this paper, we initiate the study of when efficient spectral graph theory is possible on these graphs. We investigate whether or not it is possible to solve the following problems in $n^{1+o(1)}$ time for a $\mathsf{K}$-graph $G_P$ when $d < n^{o(1)}$: $\bullet$ Multiply a given vector by the adjacency matrix or Laplacian matrix of $G_P$ $\bullet$ Find a spectral sparsifier of $G_P$ $\bullet$ Solve a Laplacian system in $G_P$'s Laplacian matrix For each of these problems, we consider all functions of the form $\mathsf{K}(u,v) = f(\|u-v\|_2^2)$ for a function $f:\mathbb{R} \rightarrow \mathbb{R}$. We provide algorithms and comparable hardness results for many such $\mathsf{K}$, including the Gaussian kernel, Neural tangent kernels, and more. For example, in dimension $d = \Omega(\log n)$, we show that there is a parameter associated with the function $f$ for which low parameter values imply $n^{1+o(1)}$ time algorithms for all three of these problems and high parameter values imply the nonexistence of subquadratic time algorithms assuming Strong Exponential Time Hypothesis ($\mathsf{SETH}$), given natural assumptions on $f$. As part of our results, we also show that the exponential dependence on the dimension $d$ in the celebrated fast multipole method of Greengard and Rokhlin cannot be improved, assuming $\mathsf{SETH}$, for a broad class of functions $f$. To the best of our knowledge, this is the first formal limitation proven about fast multipole methods.

The loss surface of an overparameterized neural network (NN) possesses many global minima of zero training error. We explain how common variants of the standard NN training procedure change the minimizer obtained. First, we make explicit how the size of the initialization of a strongly overparameterized NN affects the minimizer and can deteriorate its final test performance. We propose a strategy to limit this effect. Then, we demonstrate that for adaptive optimization such as AdaGrad, the obtained minimizer generally differs from the gradient descent (GD) minimizer. This adaptive minimizer is changed further by stochastic mini-batch training, even though in the non-adaptive case GD and stochastic GD result in essentially the same minimizer. Lastly, we explain that these effects remain relevant for less overparameterized NNs. While overparameterization has its benefits, our work highlights that it induces sources of error absent from underparameterized models, some of which can be challenging to control.

Learned optimizers are algorithms that can themselves be trained to solve optimization problems. In contrast to baseline optimizers (such as momentum or Adam) that use simple update rules derived from theoretical principles, learned optimizers use flexible, high-dimensional, nonlinear parameterizations. Although this can lead to better performance in certain settings, their inner workings remain a mystery. How is a learned optimizer able to outperform a well tuned baseline? Has it learned a sophisticated combination of existing optimization techniques, or is it implementing completely new behavior? In this work, we address these questions by careful analysis and visualization of learned optimizers. We study learned optimizers trained from scratch on three disparate tasks, and discover that they have learned interpretable mechanisms, including: momentum, gradient clipping, learning rate schedules, and a new form of learning rate adaptation. Moreover, we show how the dynamics of learned optimizers enables these behaviors. Our results help elucidate the previously murky understanding of how learned optimizers work, and establish tools for interpreting future learned optimizers.

Neural networks have been achieving high generalization performance on many tasks despite being highly over-parameterized. Since classical statistical learning theory struggles to explain this behavior, much effort has recently been focused on uncovering the mechanisms behind it, in the hope of developing a more adequate theoretical framework and having a better control over the trained models. In this work, we adopt an alternate perspective, viewing the neural network as a dynamical system displacing input particles over time. We conduct a series of experiments and, by analyzing the network's behavior through its displacements, we show the presence of a low kinetic energy displacement bias in the transport map of the network, and link this bias with generalization performance. From this observation, we reformulate the learning problem as follows: finding neural networks which solve the task while transporting the data as efficiently as possible. This offers a novel formulation of the learning problem which allows us to provide regularity results for the solution network, based on Optimal Transport theory. From a practical viewpoint, this allows us to propose a new learning algorithm, which automatically adapts to the complexity of the given task, and leads to networks with a high generalization ability even in low data regimes.

We initiate the study of fair classifiers that are robust to perturbations in the training distribution. Despite recent progress, the literature on fairness has largely ignored the design of fair and robust classifiers. In this work, we develop classifiers that are fair not only with respect to the training distribution, but also for a class of distributions that are weighted perturbations of the training samples. We formulate a min-max objective function whose goal is to minimize a distributionally robust training loss, and at the same time, find a classifier that is fair with respect to a class of distributions. We first reduce this problem to finding a fair classifier that is robust with respect to the class of distributions. Based on online learning algorithm, we develop an iterative algorithm that provably converges to such a fair and robust solution. Experiments on standard machine learning fairness datasets suggest that, compared to the state-of-the-art fair classifiers, our classifier retains fairness guarantees and test accuracy for a large class of perturbations on the test set. Furthermore, our experiments show that there is an inherent trade-off between fairness robustness and accuracy of such classifiers.

Autonomous robot-assisted feeding requires the ability to acquire a wide variety of food items. However, it is impossible for such a system to be trained on all types of food in existence. Therefore, a key challenge is choosing a manipulation strategy for a previously unseen food item. Previous work showed that the problem can be represented as a linear contextual bandit on visual information. However, food has a wide variety of multi-modal properties relevant to manipulation that can be hard to distinguish visually. Our key insight is that we can leverage the haptic information we collect during manipulation to learn some of these properties and more quickly adapt our visual model to previously unseen food. In general, we propose a modified linear contextual bandit framework augmented with post hoc context observed after action selection to empirically increase learning speed (as measured by cross-validation mean square error) and reduce cumulative regret. Experiments on synthetic data demonstrate that this effect is more pronounced when the dimensionality of the context is large relative to the post hoc context or when the post hoc context model is particularly easy to learn. Finally, we apply this framework to the bite acquisition problem and demonstrate the acquisition of 8 previously unseen types of food with 21% fewer failures across 64 attempts.