Non-parallel many-to-many voice conversion remains an interesting but challenging speech processing task. Recently, AutoVC, a conditional autoencoder based method, achieved excellent conversion results by disentangling the speaker identity and the speech content using information-constraining bottlenecks. However, due to the pure autoencoder training method, it is difficult to evaluate the separation effect of content and speaker identity. In this paper, a novel voice conversion framework, named $\boldsymbol T$ext $\boldsymbol G$uided $\boldsymbol A$utoVC(TGAVC), is proposed to more effectively separate content and timbre from speech, where an expected content embedding produced based on the text transcriptions is designed to guide the extraction of voice content. In addition, the adversarial training is applied to eliminate the speaker identity information in the estimated content embedding extracted from speech. Under the guidance of the expected content embedding and the adversarial training, the content encoder is trained to extract speaker-independent content embedding from speech. Experiments on AIShell-3 dataset show that the proposed model outperforms AutoVC in terms of naturalness and similarity of converted speech.

Deep and wide neural networks successfully fit very complex functions today, but dense models are starting to be prohibitively expensive for inference. To mitigate this, one promising direction is networks that activate a sparse subgraph of the network. The subgraph is chosen by a data-dependent routing function, enforcing a fixed mapping of inputs to subnetworks (e.g., the Mixture of Experts (MoE) paradigm in Switch Transformers). However, prior work is largely empirical, and while existing routing functions work well in practice, they do not lead to theoretical guarantees on approximation ability. We aim to provide a theoretical explanation for the power of sparse networks. As our first contribution, we present a formal model of data-dependent sparse networks that captures salient aspects of popular architectures. We then introduce a routing function based on locality sensitive hashing (LSH) that enables us to reason about how well sparse networks approximate target functions. After representing LSH-based sparse networks with our model, we prove that sparse networks can match the approximation power of dense networks on Lipschitz functions. Applying LSH on the input vectors means that the experts interpolate the target function in different subregions of the input space. To support our theory, we define various datasets based on Lipschitz target functions, and we show that sparse networks give a favorable trade-off between number of active units and approximation quality.

Clinical word embeddings are extensively used in various Bio-NLP problems as a state-of-the-art feature vector representation. Although they are quite successful at the semantic representation of words, due to the dataset - which potentially carries statistical and societal bias - on which they are trained, they might exhibit gender stereotypes. This study analyses gender bias of clinical embeddings on three medical categories: mental disorders, sexually transmitted diseases, and personality traits. To this extent, we analyze two different pre-trained embeddings namely (contextualized) clinical-BERT and (non-contextualized) BioWordVec. We show that both embeddings are biased towards sensitive gender groups but BioWordVec exhibits a higher bias than clinical-BERT for all three categories. Moreover, our analyses show that clinical embeddings carry a high degree of bias for some medical terms and diseases which is conflicting with medical literature. Having such an ill-founded relationship might cause harm in downstream applications that use clinical embeddings.

This week's list of top data news highlights covers July 30, 2022 to August 5, 2022 and includes articles on practicing complex surgeries with virtual reality and using an AI system to create an advertising campaign. India's National Tiger Conservation Authority (NCTA) has used over 26,000 cameras to capture over 24 million images of tigers around the country. Conservationists are using an AI system to identify tigers found in the images and quantify the total tiger population in the country. The NCTA plans to use an AI system to map patrol routes throughout sanctuaries to better monitor tigers next. Researchers at the University of New Orleans, Louisiana Department of Environmental Quality, and Jefferson Parish Department of Environmental Affairs have used a supercomputer to simulate the diffusion and dispersion of chemical compounds that can deodorize a landfill.

Biclustering algorithms partition data and covariates simultaneously, providing new insights in several domains, such as analyzing gene expression to discover new biological functions. This paper develops a new model-free biclustering algorithm in abstract spaces using the notions of energy distance (ED) and the maximum mean discrepancy (MMD) -- two distances between probability distributions capable of handling complex data such as curves or graphs. The proposed method can learn more general and complex cluster shapes than most existing literature approaches, which usually focus on detecting mean and variance differences. Although the biclustering configurations of our approach are constrained to create disjoint structures at the datum and covariate levels, the results are competitive. Our results are similar to state-of-the-art methods in their optimal scenarios, assuming a proper kernel choice, outperforming them when cluster differences are concentrated in higher-order moments. The model's performance has been tested in several situations that involve simulated and real-world datasets. Finally, new theoretical consistency results are established using some tools of the theory of optimal transport.

We propose and assess an alternative quantum generator architecture in the context of generative adversarial learning for Monte Carlo event generation, used to simulate particle physics processes at the Large Hadron Collider (LHC). We validate this methodology by implementing the quantum network on artificial data generated from known underlying distributions. The network is then applied to Monte Carlo-generated datasets of specific LHC scattering processes. The new quantum generator architecture leads to a generalization of the state-of-the-art implementations, achieving smaller Kullback-Leibler divergences even with shallow-depth networks. Moreover, the quantum generator successfully learns the underlying distribution functions even if trained with small training sample sets; this is particularly interesting for data augmentation applications. We deploy this novel methodology on two different quantum hardware architectures, trapped-ion and superconducting technologies, to test its hardware-independent viability.

This paper introduces a system that learns to sing new tunes by listening to examples. It extracts sequencing rules from input music and uses these rules to generate new tunes, which are sung by a vocal synthesiser. We developed a method to represent rules for musical composition as quantum circuits. We claim that such musical rules are quantum native: they are naturally encodable in the amplitudes of quantum states. To evaluate a rule to generate a subsequent event, the system builds the respective quantum circuit dynamically and measures it. After a brief discussion about the vocal synthesis methods that we have been experimenting with, the paper introduces our novel generative music method through a practical example. The paper shows some experiments and concludes with a discussion about harnessing the creative potential of the system.

Current deep learning solutions are well known for not informing whether they can reliably classify an example during inference. One of the most effective ways to build more reliable deep learning solutions is to improve their performance in the so-called out-of-distribution detection task, which essentially consists of "know that you do not know" or "know the unknown". In other words, out-of-distribution detection capable systems may reject performing a nonsense classification when submitted to instances of classes on which the neural network was not trained. This thesis tackles the defiant out-of-distribution detection task by proposing novel loss functions and detection scores. Uncertainty estimation is also a crucial auxiliary task in building more robust deep learning systems. Therefore, we also deal with this robustness-related task, which evaluates how realistic the probabilities presented by the deep neural network are. To demonstrate the effectiveness of our approach, in addition to a substantial set of experiments, which includes state-of-the-art results, we use arguments based on the principle of maximum entropy to establish the theoretical foundation of the proposed approaches. Unlike most current methods, our losses and scores are seamless and principled solutions that produce accurate predictions in addition to fast and efficient inference. Moreover, our approaches can be incorporated into current and future projects simply by replacing the loss used to train the deep neural network and computing a rapid score for detection.

The importance of building text-to-SQL parsers which can be applied to new databases has long been acknowledged, and a critical step to achieve this goal is schema linking, i.e., properly recognizing mentions of unseen columns or tables when generating SQLs. In this work, we propose a novel framework to elicit relational structures from large-scale pre-trained language models (PLMs) via a probing procedure based on Poincar\'e distance metric, and use the induced relations to augment current graph-based parsers for better schema linking. Compared with commonly-used rule-based methods for schema linking, we found that probing relations can robustly capture semantic correspondences, even when surface forms of mentions and entities differ. Moreover, our probing procedure is entirely unsupervised and requires no additional parameters. Extensive experiments show that our framework sets new state-of-the-art performance on three benchmarks. We empirically verify that our probing procedure can indeed find desired relational structures through qualitative analysis. Our code can be found at https://github.com/AlibabaResearch/DAMO-ConvAI.

Avoiding congestion and controlling traffic in urban scenarios is becoming nowadays of paramount importance due to the rapid growth of our cities' population and vehicles. The effective control of urban traffic as a means to mitigate congestion can be beneficial in an economic, environmental and health way. In this paper, a framework which allows to efficiently simulate and optimize traffic flow in a large roads' network with hundreds of vehicles is presented. The framework leverages on an Answer Set Programming (ASP) encoding to formally describe the movements of vehicles inside a network. Taking advantage of the ability to specify optimization constraints in ASP and the off-the-shelf solver Clingo, it is then possible to optimize the routes of vehicles inside the network to reduce a range of relevant metrics (e.g., travel times or emissions). Finally, an analysis on real-world traffic data is performed, utilizing the state-of-the-art Urban Mobility Simulator (SUMO) to keep track of the state of the network, test the correctness of the solution and to prove the efficiency and capabilities of the presented solution.