The text-to-SQL task is an active challenge in Natural Language Processing. Many existing solutions focus on using black-box language models extended with specialized components within customized end-to-end text-to-SQL pipelines. While these solutions use both closed-source proprietary language models and coding-oriented open-source models, there is a lack of research regarding SQL-specific small generative models. At the same time, recent advancements in self-correcting generation strategies show promise for improving the capabilities of existing architectures. The application of these concepts to the text-to-SQL task remains unexplored. In this paper, we introduce RetrySQL, a new approach to training text-to-SQL generation models. We prepare reasoning steps for reference SQL queries and then corrupt them to create retry data that contains both incorrect and corrected steps, divided with a special token. We continuously pre-train open-source coding models with this data and demonstrate that retry steps yield an improvements of up to 4 and 9 percentage points for overall and challenging execution metrics, respectively, as compared to pre-training without retry data. We showcase that the self-correcting behavior is learned by the model and the increase in downstream accuracy metrics is a result of this additional skill. Finally, we incorporate RetrySQL-trained models into the full text-to-SQL pipeline and showcase that they are competitive in terms of execution accuracy with proprietary models that contain orders of magnitude more parameters. RetrySQL demonstrates that self-correction can be learned in the text-to-SQL task and provides a novel way of improving generation accuracy for small SQL-oriented language models.

In recent years, many approaches have been proposed for automated heart sound analysis. However, most of these algorithms separate the two steps required for an accurate detection of abnormal cardiac sounds: segmentation of stethoscope recording into individual heart cycles and detection of heart murmurs. In this work we propose a method to train a neural network to perform both of these tasks simultaneously achieving a synergy effect. Despite the fact that the method uses both types of labels for training, they don't have to be specified for all training examples. Moreover it also supports negative examples, i.e recordings with no heart sounds. The result is a single neural network model that can detect individual heartbeat cycles, segment these into the 4 heartbeat phases and predict heart murmur presence. This is achieved by using a training loss function that incorporates relations between the different output types and uses masking in case of missing labels. We evaluated our results on the popular 2016 PhysioNet/CinC Challenge dataset for heart sounds and benchmarked it with respect to three state-of-the-art heart murmur detection algorithms. Our method significantly outperforms the latter algorithms achieving, in particular, an F1-score of 83.9% - an enhancement of 7.6 percentage points over the best of the considered alternatives.

To perform a precise auscultation for the purposes of examination of respiratory system normally requires the presence of an experienced doctor. With most recent advances in machine learning and artificial intelligence, automatic detection of pathological breath phenomena in sounds recorded with stethoscope becomes a reality. But to perform a full auscultation in home environment by layman is another matter, especially if the patient is a child. In this paper we propose a unique application of Reinforcement Learning for training an agent that interactively guides the end user throughout the auscultation procedure. We show that \textit{intelligent} selection of auscultation points by the agent reduces time of the examination fourfold without significant decrease in diagnosis accuracy compared to exhaustive auscultation.

In this article a DNN-based system for detection of three common voice disorders (vocal nodules, polyps and cysts; laryngeal neoplasm; unilateral vocal paralysis) is presented. The input to the algorithm is (at least 3-second long) audio recording of sustained vowel sound /a:/. The algorithm was developed as part of the "2018 FEMH Voice Data Challenge" organized by Far Eastern Memorial Hospital and obtained score value (defined in the challenge specification) of 77.44. This was the second best result before final submission. Final challenge results are not yet known during writing of this document. The document also reports changes that were made for the final submission which improved the score value in cross-validation by 0.6% points.