I was thrown a challenge by one of my colleagues – build a computer vision model that could insert any image in a video without distorting the moving object. This, as you imagine, was quite an intriguing project and I had a blast working on it. Working with videos is notoriously difficult because of their dynamic nature. Unlike images, we don't have static objects that we can easily identify and track. The complexity level goes up several levels – and that's where our hold on image processing and computer vision techniques comes to the fore.

We're releasing an API for accessing new AI models developed by OpenAI. Unlike most AI systems which are designed for one use-case, the API today provides a general-purpose "text in, text out" interface, allowing users to try it on virtually any English language task. You can now request access in order to integrate the API into your product, develop an entirely new application, or help us explore the strengths and limits of this technology. Given any text prompt, the API will return a text completion, attempting to match the pattern you gave it. You can "program" it by showing it just a few examples of what you'd like it to do; its success generally varies depending on how complex the task is.

Well here is the good news for Computer Science, Data Science, and Machine Learning Enthusiasts because Springer has released more than 70 books in Computer Science, Data Science, and Machine Learning domain and that too for free. Personally, I found the book's collection very impressive. Who can read these books? Right from the mathematics needed to kickstart your Machine Learning journey, Machine Learning basics, useful libraries, hands-on code, real-world examples, Programming in R and python, Deep Learning basics, robotics, and programming languages, etc. all are covered in these books.

Where data is available, it is desirable in geostatistical modelling to make use of additional covariates, for example terrain data, in order to improve prediction accuracy in the modelling task. While elevation itself may be important, additional explanatory power for any given problem can be sought (but not necessarily found) by filtering digital elevation models to extract higher-order derivatives such as slope angles, curvatures, and roughness. In essence, it would be beneficial to extract as much task-relevant information as possible from the elevation grid. However, given the complexities of the natural world, chance dictates that the use of 'off-the-shelf' filters is unlikely to derive covariates that provide strong explanatory power to the target variable at hand, and any attempt to manually design informative covariates is likely to be a trial-and-error process -- not optimal. In this paper we present a solution to this problem in the form of a deep learning approach to automatically deriving optimal task-specific terrain texture covariates from a standard SRTM 90m gridded digital elevation model (DEM). For our target variables we use point-sampled geochemical data from the British Geological Survey: concentrations of potassium, calcium and arsenic in stream sediments. We find that our deep learning approach produces covariates for geostatistical modelling that have surprisingly strong explanatory power on their own, with R-squared values around 0.6 for all three elements (with arsenic on the log scale). These results are achieved without the neural network being provided with easting, northing, or absolute elevation as inputs, and purely reflect the capacity of our deep neural network to extract task-specific information from terrain texture. We hope that these results will inspire further investigation into the capabilities of deep learning within geostatistical applications.

Detecting out-of-distribution (OOD) data is crucial for robust machine learning systems. Normalizing flows are flexible deep generative models that often surprisingly fail to distinguish between in- and out-of-distribution data: a flow trained on pictures of clothing assigns higher likelihood to handwritten digits. We investigate why normalizing flows perform poorly for OOD detection. We demonstrate that flows learn local pixel correlations and generic image-to-latent-space transformations which are not specific to the target image dataset. We show that by modifying the architecture of flow coupling layers we can bias the flow towards learning the semantic structure of the target data, improving OOD detection. Our investigation reveals that properties that enable flows to generate high-fidelity images can have a detrimental effect on OOD detection.

Over the past few years, the advancements of deep artificial neural networks (ANNs) have led to remarkable success in various cognitive tasks (e.g., vision, language and behavior). In some cases, neural networks have outperformed the conventional algorithms and achieved human-level performance [1, 2]. However, recent ANNs are becoming extremely compute-intensive and often do not generalize well to previously unseen data during training. On the other hand, human brain can reliably learn and compute intricate cognitive tasks with only a few watts of power budget. Recently, Spiking Neural Networks (SNNs) have been explored toward realizing robust and energy-efficient machine intelligence guided by the cues from neuroscience experiments [3]. SNNs are categorized as the new generation neural networks [4] based on their neuronal functionalities. A variety of spiking neuron models largely resemble biological neuronal mechanisms, which transmit information through discrete spatiotemporal events (or spikes). These spiking neuron models can be characterized by their internal state called the membrane potential. A spiking neuron integrates the inputs over time and fires a spike-output whenever the membrane potential exceeds a threshold.

Fast and stable fluid simulations are an essential prerequisite for applications ranging from computer aided aerodynamic design of automobiles or airplanes to simulations of physical effects in CGI to research in meteorology. Recent differentiable fluid simulations allow gradient based methods to optimize e.g. fluid control systems in an informed manner. Solving the partial differential equations governed by the dynamics of the underlying physical systems, however, is a challenging task and current numerical approximation schemes still come at high computational costs. In this work, we propose an unsupervised framework that allows powerful deep neural networks to learn the dynamics of incompressible fluids end to end on a grid-based representation. For this purpose, we introduce a loss function that penalizes residuals of the incompressible Navier Stokes equations. After training, the framework yields models that are capable of fast and differentiable fluid simulations and can handle various fluid phenomena such as the Magnus effect and K\'arm\'an vortex streets. Besides demonstrating its real-time capability on a GPU, we exploit our approach in a control optimization scenario.

Large neural network models present a hefty communication challenge to distributed Stochastic Gradient Descent (SGD), with a communication complexity of O(n) per worker for a model of n parameters. Many sparsification and quantization techniques have been proposed to compress the gradients, some reducing the communication complexity to O(k), where k << n. In this paper, we introduce a strategy called two-level gradient averaging (A2SGD) to consolidate all gradients down to merely two local averages per worker before the computation of two global averages for an updated model. A2SGD also retains local errors to maintain the variance for fast convergence. Our theoretical analysis shows that A2SGD converges similarly like the default distributed SGD algorithm. Our evaluation validates the theoretical conclusion and demonstrates that A2SGD significantly reduces the communication traffic per worker, and improves the overall training time of LSTM-PTB by 3.2x and 23.2x, respectively, compared to Top-K and QSGD. To the best of our knowledge, A2SGD is the first to achieve O(1) communication complexity per worker for distributed SGD.

The estimation of the ratio of two probability densities has garnered attention as the density ratio is useful in various machine learning tasks, such as anomaly detection and domain adaptation. To estimate the density ratio, methods collectively known as direct density ratio estimation (DRE) have been explored. These methods are based on the minimization of the Bregman (BR) divergence between a density ratio model and the true density ratio. However, existing direct DRE suffers from serious overfitting when using flexible models such as neural networks. In this paper, we introduce a non-negative correction for empirical risk using only the prior knowledge of the upper bound of the density ratio. This correction makes a DRE method more robust against overfitting and enables the use of flexible models. In the theoretical analysis, we discuss the consistency of the empirical risk. In our experiments, the proposed estimators show favorable performance in inlier-based outlier detection and covariate shift adaptation.

We present APQ for efficient deep learning inference on resource-constrained hardware. Unlike previous methods that separately search the neural architecture, pruning policy, and quantization policy, we optimize them in a joint manner. To deal with the larger design space it brings, a promising approach is to train a quantization-aware accuracy predictor to quickly get the accuracy of the quantized model and feed it to the search engine to select the best fit. However, training this quantization-aware accuracy predictor requires collecting a large number of quantized pairs, which involves quantization-aware finetuning and thus is highly time-consuming. To tackle this challenge, we propose to transfer the knowledge from a full-precision (i.e., fp32) accuracy predictor to the quantization-aware (i.e., int8) accuracy predictor, which greatly improves the sample efficiency. Besides, collecting the dataset for the fp32 accuracy predictor only requires to evaluate neural networks without any training cost by sampling from a pretrained once-for-all network, which is highly efficient. Extensive experiments on ImageNet demonstrate the benefits of our joint optimization approach. With the same accuracy, APQ reduces the latency/energy by 2x/1.3x over MobileNetV2+HAQ. Compared to the separate optimization approach (ProxylessNAS+AMC+HAQ), APQ achieves 2.3% higher ImageNet accuracy while reducing orders of magnitude GPU hours and CO2 emission, pushing the frontier for green AI that is environmental-friendly. The code and video are publicly available.