Training robust Speech to Text (STT) system require "tens of thousand" of hours of data. Variability present in the dataset, in the form of unwanted nuisances (noise) and biases (accent, gender or age) is the reason for the need of large datasets to learn general representations, which is unfeasible for low resource languages. A recently proposed deep learning approach to remove these unwanted features, called adversarial forgetting, was able to produce better results on computer vision tasks. Motivated by this, in this paper, we study the effect of de-entangling the accent information from the input speech signal on training STT systems. To this end, we use an information bottleneck architecture based on adversarial forgetting. This training scheme aims to enforce the model to learn general accent invariant speech representations. The trained STT model is tested on two unseen accents in the common voice V1. The results are in favour of STT model trained using the adversarial forgetting scheme.

The notion of concept drift refers to the phenomenon that the distribution, which is underlying the observed data, changes over time. We are interested in an identification of those features, that are most relevant for the observed drift. We distinguish between drift inducing features, for which the observed feature drift cannot be explained by any other feature, and faithfully drifting features, which correlate with the present drift of other features. This notion gives rise to minimal subsets of the feature space, which are able to characterize the observed drift as a whole. We relate this problem to the problems of feature selection and feature relevance learning, which allows us to derive a detection algorithm. We demonstrate its usefulness on different benchmarks.

We use gradient sparsification to reduce the adverse effect of differential privacy noise on performance of private machine learning models. To this aim, we employ compressed sensing and additive Laplace noise to evaluate differentially-private gradients. Noisy privacy-preserving gradients are used to perform stochastic gradient descent for training machine learning models. Sparsification, achieved by setting the smallest gradient entries to zero, can reduce the convergence speed of the training algorithm. However, by sparsification and compressed sensing, the dimension of communicated gradient and the magnitude of additive noise can be reduced. The interplay between these effects determines whether gradient sparsification improves the performance of differentially-private machine learning models. We investigate this analytically in the paper. We prove that, for small privacy budgets, compression can improve performance of privacy-preserving machine learning models. However, for large privacy budgets, compression does not necessarily improve the performance. Intuitively, this is because the effect of privacy-preserving noise is minimal in large privacy budget regime and thus improvements from gradient sparsification cannot compensate for its slower convergence.

As the world is anticipating the end of the COVID-19 pandemic, energy consumption in industry and services is likely to grow. In the longer term, the developing world will increase its energy utilization, leading to growth of global primary energy demand by of 0.4% - 0.6% per year, or a 25% increase by 2050. According to scenarios calculated by energy giant Total SE, massive electrification of transportation will lead to decarbonization, and will require a rapid growth in renewables as a source of electricity. This energy transformation will see an explosion of growth in Artificial Intelligence (AI) utilization in the sector – up 50% between 2020 and 2024 – to allow smart, 21st century grids to become the gold standard, gradually replacing the "dumb" grids laid down in the late 19th – early 20th century in Europe, North America, Japan, China and beyond. The grid is a meta-system of generation facilities, be it nuclear, gas, coal, solar, wind, and hydro, connected by high voltage wire networks to transformers, and then to sub-stations and individual buildings, households, and apartments.

We present EmoCoder, a modular encoder-decoder architecture that generalizes emotion analysis over different tasks (sentence-level, word-level, label-to-label mapping), domains (natural languages and their registers), and label formats (e.g., polarity classes, basic emotions, and affective dimensions). Experiments on 14 datasets indicate that EmoCoder learns an interpretable language-independent representation of emotions, allows seamless absorption of state-of-the-art models, and maintains strong prediction quality, even when tested on unseen combinations of domains and label formats.

This paper presents a novel alternative to Greedy Non-Maxima Suppression (NMS) in the task of bounding box selection and suppression in object detection. It proposes Confluence, an algorithm which does not rely solely on individual confidence scores to select optimal bounding boxes, nor does it rely on Intersection Over Union (IoU) to remove false positives. Using Manhattan Distance, it selects the bounding box which is closest to every other bounding box within the cluster and removes highly confluent neighboring boxes. Thus, Confluence represents a paradigm shift in bounding box selection and suppression as it is based on fundamentally different theoretical principles to Greedy NMS and its variants. Confluence is experimentally validated on RetinaNet, YOLOv3 and Mask-RCNN, using both the MS COCO and PASCAL VOC 2007 datasets. Confluence outperforms Greedy NMS in both mAP and recall on both datasets, using the challenging 0.50:0.95 mAP evaluation metric. On each detector and dataset, mAP was improved by 0.3-0.7% while recall was improved by 1.4-2.5%. A theoretical comparison of Greedy NMS and the Confluence Algorithm is provided, and quantitative results are supported by extensive qualitative results analysis. Furthermore, sensitivity analysis experiments across mAP thresholds support the conclusion that Confluence is more robust than NMS.

The paper proposes a new hybrid Bayesian network learning algorithm, termed Forward Early Dropping Hill Climbing (FEDHC), designed to work with either continuous or categorical data. FEDHC consists of a skeleton identification phase (learning the conditional associations among the variables) followed by the scoring phase that assigns the causal directions. Specifically for the case of continuous data, a robust to outliers version of FEDHC is also proposed. The paper manifests that the only implementation of MMHC in the statistical software \textit{R}, is prohibitively expensive and a new implementation is offered. The FEDHC is tested via Monte Carlo simulations that distinctly show it is computationally efficient, and produces Bayesian networks of similar to, or of higher accuracy than MMHC and PCHC. FEDHC yields more accurate Bayesian networks than PCHC with continuous data but less accurate with categorical data. Finally, an application of FEDHC, PCHC and MMHC algorithms to real data, from the field of economics, is demonstrated using the statistical software \textit{R}.

There are many different causal effect estimators in causal inference. However, it is unclear how to choose between these estimators because there is no ground-truth for causal effects. A commonly used option is to simulate synthetic data, where the ground-truth is known. However, the best causal estimators on synthetic data are unlikely to be the best causal estimators on realistic data. An ideal benchmark for causal estimators would both (a) yield ground-truth values of the causal effects and (b) be representative of real data. Using flexible generative models, we provide a benchmark that both yields ground-truth and is realistic. Using this benchmark, we evaluate 66 different causal estimators.

The prototype technology brings together imaging, processing, machine learning and memory in one electronic chip, powered by light. If scientists could train artificial intelligence (AI) systems to remember the images they capture from their photodetectors and learn from them--all in one package--it would be one step closer to an artificial brain. But the huge data sets and computer power required to make sense of them usually require offloading images elsewhere for processing--unlike a natural brain. Now, researchers in Australia have developed a neuromorphic imaging chip that performs image pre-processing and recognition by itself (Adv. The optically driven chip, made of two-dimensional black phosphorus, demonstrates a way to combine AI software and imaging hardware in a brain-like package that could run autonomously.

As some of you know, we have had a Flipboard magazine that deals with Blockchain, Cryptocurrencies, and of course of Fintech for quite some time. We are known for publishing the short blog post - Crypto stories you may have missed. We have taken a leaf from our book and done the same with artificial intelligence, machine learning, and robotics. From this moment on, we will be putting out a similar blog post for AI, ML, and robotics. We intend to have this continue into the near future.