Strong empirical evidence that one machine-learning algorithm A outperforms another one B ideally calls for multiple trials optimizing the learning pipeline over sources of variation such as data sampling, data augmentation, parameter initialization, and hyperparameters choices. This is prohibitively expensive, and corners are cut to reach conclusions. We model the whole benchmarking process, revealing that variance due to data sampling, parameter initialization and hyperparameter choice impact markedly the results. We analyze the predominant comparison methods used today in the light of this variance. We show a counter-intuitive result that adding more sources of variation to an imperfect estimator approaches better the ideal estimator at a 51 times reduction in compute cost. Building on these results, we study the error rate of detecting improvements, on five different deep-learning tasks/architectures. This study leads us to propose recommendations for performance comparisons.

We study the cocktail party problem and propose a novel attention network called Tune-In, abbreviated for training under negative environments with interference. It firstly learns two separate spaces of speaker-knowledge and speech-stimuli based on a shared feature space, where a new block structure is designed as the building block for all spaces, and then cooperatively solves different tasks. Between the two spaces, information is cast towards each other via a novel cross- and dual-attention mechanism, mimicking the bottom-up and top-down processes of a human's cocktail party effect. It turns out that substantially discriminative and generalizable speaker representations can be learnt in severely interfered conditions via our self-supervised training. The experimental results verify this seeming paradox. The learnt speaker embedding has superior discriminative power than a standard speaker verification method; meanwhile, Tune-In achieves remarkably better speech separation performances in terms of SI-SNRi and SDRi consistently in all test modes, and especially at lower memory and computational consumption, than state-of-the-art benchmark systems.

Ceramic insulators are important to electronic systems, designed and installed to protect humans from the danger of high voltage electric current. However, insulators are not immortal, and natural deterioration can gradually damage them. Therefore, the condition of insulators must be continually monitored, which is normally done using UAVs. UAVs collect many images of insulators, and these images are then analyzed to identify those that are damaged. Here we describe AdeNet as a deep neural network designed to identify damaged insulators, and test multiple approaches to automatic analysis of the condition of insulators. Several deep neural networks were tested, as were shallow learning methods. The best results (88.8%) were achieved using AdeNet without transfer learning. AdeNet also reduced the false negative rate to 7%. While the method cannot fully replace human inspection, its high throughput can reduce the amount of labor required to monitor lines for damaged insulators and provide early warning to replace damaged insulators.

Robust and accurate detection of small moving targets in cluttered moving backgrounds is a significant and challenging problem for robotic visual systems to perform search and tracking tasks. Inspired by the neural circuitry of elementary motion vision in the mammalian retina, this paper proposes a bioinspired retinal neural network based on a new neurodynamics-based temporal filtering and multiform 2-D spatial Gabor filtering. This model can estimate motion direction accurately via only two perpendicular spatiotemporal filtering signals, and respond to small targets of different sizes and velocities by adjusting the dendrite field size of the spatial filter. Meanwhile, an algorithm of directionally selective inhibition is proposed to suppress the target-like features in the moving background, which can reduce the influence of background motion effectively. Extensive synthetic and real-data experiments show that the proposed model works stably for small targets of a wider size and velocity range, and has better detection performance than other bioinspired models. Additionally, it can also extract the information of motion direction and motion energy accurately and rapidly.

Q-learning (QL), a common reinforcement learning algorithm, suffers from over-estimation bias due to the maximization term in the optimal Bellman operator. This bias may lead to sub-optimal behavior. Double-Q-learning tackles this issue by utilizing two estimators, yet results in an under-estimation bias. Similar to over-estimation in Q-learning, in certain scenarios, the under-estimation bias may degrade performance. In this work, we introduce a new bias-reduced algorithm called Ensemble Bootstrapped Q-Learning (EBQL), a natural extension of Double-Q-learning to ensembles. We analyze our method both theoretically and empirically. Theoretically, we prove that EBQL-like updates yield lower MSE when estimating the maximal mean of a set of independent random variables. Empirically, we show that there exist domains where both over and under-estimation result in sub-optimal performance. Finally, We demonstrate the superior performance of a deep RL variant of EBQL over other deep QL algorithms for a suite of ATARI games.

Federated learning becomes a prominent approach when different entities want to learn collaboratively a common model without sharing their training data. However, Federated learning has two main drawbacks. First, it is quite bandwidth inefficient as it involves a lot of message exchanges between the aggregating server and the participating entities. This bandwidth and corresponding processing costs could be prohibitive if the participating entities are, for example, mobile devices. Furthermore, although federated learning improves privacy by not sharing data, recent attacks have shown that it still leaks information about the training data. This paper presents a novel privacy-preserving federated learning scheme. The proposed scheme provides theoretical privacy guarantees, as it is based on Differential Privacy. Furthermore, it optimizes the model accuracy by constraining the model learning phase on few selected weights. Finally, as shown experimentally, it reduces the upstream and downstream bandwidth by up to 99.9% compared to standard federated learning, making it practical for mobile systems.

Naive Bayes is a classification algorithm for binary (two-class) and multiclass classification problems. It is called Naive Bayes or idiot Bayes because the calculations of the probabilities for each class are simplified to make their calculations tractable. Rather than attempting to calculate the probabilities of each attribute value, they are assumed to be conditionally independent given the class value. This is a very strong assumption that is most unlikely in real data, i.e. that the attributes do not interact. Nevertheless, the approach performs surprisingly well on data where this assumption does not hold.

The Varroa destructor mite is one of the most dangerous Honey Bee (Apis mellifera) parasites worldwide and the bee colonies have to be regularly monitored in order to control its spread. Here we present an object detector based method for health state monitoring of bee colonies. This method has the potential for online measurement and processing. In our experiment, we compare the YOLO and SSD object detectors along with the Deep SVDD anomaly detector. Based on the custom dataset with 600 ground-truth images of healthy and infected bees in various scenes, the detectors reached a high F1 score up to 0.874 in the infected bee detection and up to 0.727 in the detection of the Varroa Destructor mite itself. The results demonstrate the potential of this approach, which will be later used in the real-time computer vision based honey bee inspection system. To the best of our knowledge, this study is the first one using object detectors for this purpose. We expect that performance of those object detectors will enable us to inspect the health status of the honey bee colonies.

We re-examine the topic of machine-learned clause selection guidance in saturation-based theorem provers. The central idea, recently popularized by the ENIGMA system, is to learn a classifier for recognizing clauses that appeared in previously discovered proofs. In subsequent runs, clauses classified positively are prioritized for selection. We propose several improvements to this approach and experimentally confirm their viability. For the demonstration, we use a Recursive Neural Network to classify clauses based on their derivation history and the presence or absence of automatically supplied theory axioms therein. The automatic theorem prover Vampire guided by the network achieves a 41 % improvement on a relevant subset of smt-lib in a real time evaluation.

Commonly, Deep Neural Networks (DNNs) generalize well on samples drawn from a distribution similar to that of the training set. However, DNNs' predictions are brittle and unreliable when the test samples are drawn from a dissimilar distribution. This presents a major concern for deployment in real-world applications, where such behavior may come at a great cost -- as in the case of autonomous vehicles or healthcare applications. This paper frames the Out Of Distribution (OOD) detection problem in DNN as a statistical hypothesis testing problem. Unlike previous OOD detection heuristics, our framework is guaranteed to maintain the false positive rate (detecting OOD as in-distribution) for test data. We build on this framework to suggest a novel OOD procedure based on low-order statistics. Our method achieves comparable or better than state-of-the-art results on well-accepted OOD benchmarks without retraining the network parameters -- and at a fraction of the computational cost.