Neural audio synthesis is an actively researched topic, having yielded a wide range of techniques that leverages machine learning architectures. Google Magenta elaborated a novel approach called Differential Digital Signal Processing (DDSP) that incorporates deep neural networks with preconditioned digital signal processing techniques, reaching state-of-the-art results especially in timbre transfer applications. However, most of these techniques, including the DDSP, are generally not applicable in real-time constraints, making them ineligible in a musical workflow. In this paper, we present a real-time implementation of the DDSP library embedded in a virtual synthesizer as a plug-in that can be used in a Digital Audio Workstation. We focused on timbre transfer from learned representations of real instruments to arbitrary sound inputs as well as controlling these models by MIDI. Furthermore, we developed a GUI for intuitive high-level controls which can be used for post-processing and manipulating the parameters estimated by the neural network. We have conducted a user experience test with seven participants online. The results indicated that our users found the interface appealing, easy to understand, and worth exploring further. At the same time, we have identified issues in the timbre transfer quality, in some components we did not implement, and in installation and distribution of our plugin. The next iteration of our design will address these issues. Our real-time MATLAB and JUCE implementations are available at https://github.com/SMC704/juce-ddsp and https://github.com/SMC704/matlab-ddsp , respectively.
We show that a purely tactile dextrous in-hand manipulation task with continuous regrasping, requiring permanent force closure, can be learned from scratch and executed robustly on a torque-controlled humanoid robotic hand. The task is rotating a cube without dropping it, but in contrast to OpenAI's seminal cube manipulation task, the palm faces downwards and no cameras but only the hand's position and torque sensing are used. Although the task seems simple, it combines for the first time all the challenges in execution as well as learning that are important for using in-hand manipulation in real-world applications. We efficiently train in a precisely modeled and identified rigid body simulation with off-policy deep reinforcement learning, significantly sped up by a domain adapted curriculum, leading to a moderate 600 CPU hours of training time. The resulting policy is robustly transferred to the real humanoid DLR Hand-II, e.g., reaching more than 46 full 2${\pi}$ rotations of the cube in a single run and allowing for disturbances like different cube sizes, hand orientation, or pulling a finger.
Despite its importance, unsupervised domain adaptation (UDA) on LiDAR semantic segmentation is a task that has not received much attention from the research community. Only recently, a completion-based 3D method has been proposed to tackle the problem and formally set up the adaptive scenarios. However, the proposed pipeline is complex, voxel-based and requires multi-stage inference, which inhibits it for real-time inference. We propose a range image-based, effective and efficient method for solving UDA on LiDAR segmentation. The method exploits class prototypes from the source domain to pseudo label target domain pixels, which is a research direction showing good performance in UDA for natural image semantic segmentation. Applying such approaches to LiDAR scans has not been considered because of the severe domain shift and lack of pre-trained feature extractor that is unavailable in the LiDAR segmentation setup. However, we show that proper strategies, including reconstruction-based pre-training, enhanced prototypes, and selective pseudo labeling based on distance to prototypes, is sufficient enough to enable the use of prototypical approaches. We evaluate the performance of our method on the recently proposed LiDAR segmentation UDA scenarios. Our method achieves remarkable performance among contemporary methods.
We develop an algorithm for parameter-free stochastic convex optimization (SCO) whose rate of convergence is only a double-logarithmic factor larger than the optimal rate for the corresponding known-parameter setting. In contrast, the best previously known rates for parameter-free SCO are based on online parameter-free regret bounds, which contain unavoidable excess logarithmic terms compared to their known-parameter counterparts. Our algorithm is conceptually simple, has high-probability guarantees, and is also partially adaptive to unknown gradient norms, smoothness, and strong convexity. At the heart of our results is a novel parameter-free certificate for SGD step size choice, and a time-uniform concentration result that assumes no a-priori bounds on SGD iterates.
The two most popular loss functions for streaming end-to-end automatic speech recognition (ASR) are the RNN-Transducer (RNN-T) and the connectionist temporal classification (CTC) objectives. Both perform an alignment-free training by marginalizing over all possible alignments, but use different transition rules. Between these two loss types we can classify the monotonic RNN-T (MonoRNN-T) and the recently proposed CTC-like Transducer (CTC-T), which both can be realized using the graph temporal classification-transducer (GTC-T) loss function. Monotonic transducers have a few advantages. First, RNN-T can suffer from runaway hallucination, where a model keeps emitting non-blank symbols without advancing in time, often in an infinite loop. Secondly, monotonic transducers consume exactly one model score per time step and are therefore more compatible and unifiable with traditional FST-based hybrid ASR decoders. However, the MonoRNN-T so far has been found to have worse accuracy than RNN-T. It does not have to be that way, though: By regularizing the training - via joint LAS training or parameter initialization from RNN-T - both MonoRNN-T and CTC-T perform as well - or better - than RNN-T. This is demonstrated for LibriSpeech and for a large-scale in-house data set.
Delirium occurs in about 80% cases in the Intensive Care Unit (ICU) and is associated with a longer hospital stay, increased mortality and other related issues. Delirium does not have any biomarker-based diagnosis and is commonly treated with antipsychotic drugs (APD). However, multiple studies have shown controversy over the efficacy or safety of APD in treating delirium. Since randomized controlled trials (RCT) are costly and time-expensive, we aim to approach the research question of the efficacy of APD in the treatment of delirium using retrospective cohort analysis. We plan to use the Causal inference framework to look for the underlying causal structure model, leveraging the availability of large observational data on ICU patients. To explore safety outcomes associated with APD, we aim to build a causal model for delirium in the ICU using large observational data sets connecting various covariates correlated with delirium. We utilized the MIMIC III database, an extensive electronic health records (EHR) dataset with 53,423 distinct hospital admissions. Our null hypothesis is: there is no significant difference in outcomes for delirium patients under different drug-group in the ICU. Through our exploratory, machine learning based and causal analysis, we had findings such as: mean length-of-stay and max length-of-stay is higher for patients in Haloperidol drug group, and haloperidol group has a higher rate of death in a year compared to other two-groups. Our generated causal model explicitly shows the functional relationships between different covariates. For future work, we plan to do time-varying analysis on the dataset.
Time series prediction with neural networks have been focus of much research in the past few decades. Given the recent deep learning revolution, there has been much attention in using deep learning models for time series prediction, and hence it is important to evaluate their strengths and weaknesses. In this paper, we present an evaluation study that compares the performance of deep learning models for multi-step ahead time series prediction. Our deep learning methods compromise of simple recurrent neural networks, long short term memory (LSTM) networks, bidirectional LSTM, encoder-decoder LSTM networks, and convolutional neural networks. We also provide comparison with simple neural networks use stochastic gradient descent and adaptive gradient method (Adam) for training. We focus on univariate and multi-step-ahead prediction from benchmark time series datasets and compare with results from from the literature. The results show that bidirectional and encoder-decoder LSTM provide the best performance in accuracy for the given time series problems with different properties.
Both pedestrian and robot comfort are of the highest priority whenever a robot is placed in an environment containing human beings. In the case of pedestrian-unaware mobile robots this desire for safety leads to the freezing robot problem, where a robot confronted with a large dynamic group of obstacles (such as a crowd of pedestrians) would determine all forward navigation unsafe causing the robot to stop in place. In order to navigate in a socially compliant manner while avoiding the freezing robot problem we are interested in understanding the flow of pedestrians in crowded scenarios. By treating the pedestrians in the crowd as particles moved along by the crowd itself we can model the system as a time dependent flow field. From this flow field we can extract different flow segments that reflect the motion patterns emerging from the crowd. These motion patterns can then be accounted for during the control and navigation of a mobile robot allowing it to move safely within the flow of the crowd to reach a desired location within or beyond the flow. We combine flow-field extraction with a discrete heuristic search to create Flow-Informed path planning (FIPP). We provide empirical results showing that when compared against a trajectory-rollout local path planner, a robot using FIPP was able not only to reach its goal more quickly but also was shown to be more socially compliant than a robot using traditional techniques both in simulation and on real robots.
Large-scale language models often learn behaviors that are misaligned with user expectations. Generated text may contain offensive or toxic language, contain significant repetition, or be of a different sentiment than desired by the user. We consider the task of unlearning these misalignments by fine-tuning the language model on signals of what not to do. We introduce Quantized Reward Konditioning (Quark), an algorithm for optimizing a reward function that quantifies an (un)wanted property, while not straying too far from the original model. Quark alternates between (i) collecting samples with the current language model, (ii) sorting them into quantiles based on reward, with each quantile identified by a reward token prepended to the language model's input, and (iii) using a standard language modeling loss on samples from each quantile conditioned on its reward token, while remaining nearby the original language model via a KL-divergence penalty. By conditioning on a high-reward token at generation time, the model generates text that exhibits less of the unwanted property. For unlearning toxicity, negative sentiment, and repetition, our experiments show that Quark outperforms both strong baselines and state-of-the-art reinforcement learning methods like PPO (Schulman et al. 2017), while relying only on standard language modeling primitives.
A cornerstone of the worldwide transition to smart grids are smart meters. Smart meters typically collect and provide energy time series that are vital for various applications, such as grid simulations, fault-detection, load forecasting, load analysis, and load management. Unfortunately, these time series are often characterized by missing values that must be handled before the data can be used. A common approach to handle missing values in time series is imputation. However, existing imputation methods are designed for power time series and do not take into account the total energy of gaps, resulting in jumps or constant shifts when imputing energy time series. In order to overcome these issues, the present paper introduces the new Copy-Paste Imputation (CPI) method for energy time series. The CPI method copies data blocks with similar properties and pastes them into gaps of the time series while preserving the total energy of each gap. The new method is evaluated on a real-world dataset that contains six shares of artificially inserted missing values between 1 and 30%. It outperforms by far the three benchmark imputation methods selected for comparison. The comparison furthermore shows that the CPI method uses matching patterns and preserves the total energy of each gap while requiring only a moderate run-time.