The performance of neural network-based speech enhancement systems is primarily influenced by the model architecture, whereas training times and computational resource utilization are primarily affected by training parameters such as the batch size. Since noisy and reverberant speech mixtures can have different duration, a batching strategy is required to handle variable size inputs during training, in particular for state-of-the-art end-to-end systems. Such strategies usually strive for a compromise between zero-padding and data randomization, and can be combined with a dynamic batch size for a more consistent amount of data in each batch. However, the effect of these strategies on resource utilization and more importantly network performance is not well documented. This paper systematically investigates the effect of different batching strategies and batch sizes on the training statistics and speech enhancement performance of a Conv-TasNet, evaluated in both matched and mismatched conditions. We find that using a small batch size during training improves performance in both conditions for all batching strategies. Moreover, using sorted or bucket batching with a dynamic batch size allows for reduced training time and GPU memory usage while achieving similar performance compared to random batching with a fixed batch size.
相關內容
Large language models (LLMs) with in-context learning have demonstrated remarkable capability in the text-to-SQL task. Previous research has prompted LLMs with various demonstration-retrieval strategies and intermediate reasoning steps to enhance the performance of LLMs. However, those works often employ varied strategies when constructing the prompt text for text-to-SQL inputs, such as databases and demonstration examples. This leads to a lack of comparability in both the prompt constructions and their primary contributions. Furthermore, selecting an effective prompt construction has emerged as a persistent problem for future research. To address this limitation, we comprehensively investigate the impact of prompt constructions across various settings and provide insights for future work.
Dense retrieval has shown promise in the first-stage retrieval process when trained on in-domain labeled datasets. However, previous studies have found that dense retrieval is hard to generalize to unseen domains due to its weak modeling of domain-invariant and interpretable feature (i.e., matching signal between two texts, which is the essence of information retrieval). In this paper, we propose a novel method to improve the generalization of dense retrieval via capturing matching signal called BERM. Fully fine-grained expression and query-oriented saliency are two properties of the matching signal. Thus, in BERM, a single passage is segmented into multiple units and two unit-level requirements are proposed for representation as the constraint in training to obtain the effective matching signal. One is semantic unit balance and the other is essential matching unit extractability. Unit-level view and balanced semantics make representation express the text in a fine-grained manner. Essential matching unit extractability makes passage representation sensitive to the given query to extract the pure matching information from the passage containing complex context. Experiments on BEIR show that our method can be effectively combined with different dense retrieval training methods (vanilla, hard negatives mining and knowledge distillation) to improve its generalization ability without any additional inference overhead and target domain data.
Zero packet loss with bounded latency is necessary for many applications, such as industrial control networks, automotive Ethernet, and aircraft communication systems. Traditional networks cannot meet the such strict requirement, and thus Time-Sensitive Networking (TSN) emerges. TSN is a set of standards proposed by IEEE 802 for providing deterministic connectivity in terms of low packet loss, low packet delay variation, and guaranteed packet transport. However, to our knowledge, few existing TSN solutions can deterministically achieve zero packet loss with bounded latency. This paper fills in this blank by proposing a novel input-queueing TSN switching architecture, under which we design a TDMA-like scheduling policy (called M-TDMA) along with a sufficient condition and an EDF-like scheduling policy (called M-EDF) along with a different sufficient condition to achieve zero packet loss with bounded latency.
Hierarchical Bayesian models of perception and learning feature prominently in contemporary cognitive neuroscience where, for example, they inform computational concepts of mental disorders. This includes predictive coding and hierarchical Gaussian filtering (HGF), which differ in the nature of hierarchical representations. Predictive coding assumes that higher levels in a given hierarchy influence the state (value) of lower levels. In HGF, however, higher levels determine the rate of change at lower levels. Here, we extend the space of generative models underlying HGF to include a form of nonlinear hierarchical coupling between state values akin to predictive coding and artificial neural networks in general. We derive the update equations corresponding to this generalization of HGF and conceptualize them as connecting a network of (belief) nodes where parent nodes either predict the state of child nodes or their rate of change. This enables us to (1) create modular architectures with generic computational steps in each node of the network, and (2) disclose the hierarchical message passing implied by generalized HGF models and to compare this to comparable schemes under predictive coding. We find that the algorithmic architecture instantiated by the generalized HGF is largely compatible with that of predictive coding but extends it with some unique predictions which arise from precision and volatility related computations. Our developments enable highly flexible implementations of hierarchical Bayesian models for empirical data analysis and are available as open source software.
Phase aberration is one of the primary sources of image quality degradation in ultrasound, which is induced by spatial variations in sound speed across the heterogeneous medium. This effect disrupts transmitted waves and prevents coherent summation of echo signals, resulting in suboptimal image quality. In real experiments, obtaining non-aberrated ground truths can be extremely challenging, if not infeasible. It hinders the performance of deep learning-based phase aberration correction techniques due to sole reliance on simulated data and the presence of domain shift between simulated and experimental data. Here, for the first time, we propose a deep learning-based method that does not require reference data to compensate for the phase aberration effect. We train a network wherein both input and target output are randomly aberrated radio frequency (RF) data. Moreover, we demonstrate that a conventional loss function such as mean square error is inadequate for training the network to achieve optimal performance. Instead, we propose an adaptive mixed loss function that employs both B-mode and RF data, resulting in more efficient convergence and enhanced performance. Source code is available at \url{//code.sonography.ai}.
Continual learning for end-to-end automatic speech recognition has to contend with a number of difficulties. Fine-tuning strategies tend to lose performance on data already seen, a process known as catastrophic forgetting. On the other hand, strategies that freeze parameters and append tunable parameters must maintain multiple models. We suggest a strategy that maintains only a single model for inference and avoids catastrophic forgetting. Our experiments show that a simple linear interpolation of several models' parameters, each fine-tuned from the same generalist model, results in a single model that performs well on all tested data. For our experiments we selected two open-source end-to-end speech recognition models pre-trained on large datasets and fine-tuned them on 3 separate datasets: SGPISpeech, CORAAL, and DiPCo. The proposed average of domain experts model performs well on all tested data, and has almost no loss in performance on data from the domain of original training.
The past few years have seen rapid progress in combining reinforcement learning (RL) with deep learning. Various breakthroughs ranging from games to robotics have spurred the interest in designing sophisticated RL algorithms and systems. However, the prevailing workflow in RL is to learn tabula rasa, which may incur computational inefficiency. This precludes continuous deployment of RL algorithms and potentially excludes researchers without large-scale computing resources. In many other areas of machine learning, the pretraining paradigm has shown to be effective in acquiring transferable knowledge, which can be utilized for a variety of downstream tasks. Recently, we saw a surge of interest in Pretraining for Deep RL with promising results. However, much of the research has been based on different experimental settings. Due to the nature of RL, pretraining in this field is faced with unique challenges and hence requires new design principles. In this survey, we seek to systematically review existing works in pretraining for deep reinforcement learning, provide a taxonomy of these methods, discuss each sub-field, and bring attention to open problems and future directions.
We hypothesize that due to the greedy nature of learning in multi-modal deep neural networks, these models tend to rely on just one modality while under-fitting the other modalities. Such behavior is counter-intuitive and hurts the models' generalization, as we observe empirically. To estimate the model's dependence on each modality, we compute the gain on the accuracy when the model has access to it in addition to another modality. We refer to this gain as the conditional utilization rate. In the experiments, we consistently observe an imbalance in conditional utilization rates between modalities, across multiple tasks and architectures. Since conditional utilization rate cannot be computed efficiently during training, we introduce a proxy for it based on the pace at which the model learns from each modality, which we refer to as the conditional learning speed. We propose an algorithm to balance the conditional learning speeds between modalities during training and demonstrate that it indeed addresses the issue of greedy learning. The proposed algorithm improves the model's generalization on three datasets: Colored MNIST, Princeton ModelNet40, and NVIDIA Dynamic Hand Gesture.
The growing energy and performance costs of deep learning have driven the community to reduce the size of neural networks by selectively pruning components. Similarly to their biological counterparts, sparse networks generalize just as well, if not better than, the original dense networks. Sparsity can reduce the memory footprint of regular networks to fit mobile devices, as well as shorten training time for ever growing networks. In this paper, we survey prior work on sparsity in deep learning and provide an extensive tutorial of sparsification for both inference and training. We describe approaches to remove and add elements of neural networks, different training strategies to achieve model sparsity, and mechanisms to exploit sparsity in practice. Our work distills ideas from more than 300 research papers and provides guidance to practitioners who wish to utilize sparsity today, as well as to researchers whose goal is to push the frontier forward. We include the necessary background on mathematical methods in sparsification, describe phenomena such as early structure adaptation, the intricate relations between sparsity and the training process, and show techniques for achieving acceleration on real hardware. We also define a metric of pruned parameter efficiency that could serve as a baseline for comparison of different sparse networks. We close by speculating on how sparsity can improve future workloads and outline major open problems in the field.
The Q-learning algorithm is known to be affected by the maximization bias, i.e. the systematic overestimation of action values, an important issue that has recently received renewed attention. Double Q-learning has been proposed as an efficient algorithm to mitigate this bias. However, this comes at the price of an underestimation of action values, in addition to increased memory requirements and a slower convergence. In this paper, we introduce a new way to address the maximization bias in the form of a "self-correcting algorithm" for approximating the maximum of an expected value. Our method balances the overestimation of the single estimator used in conventional Q-learning and the underestimation of the double estimator used in Double Q-learning. Applying this strategy to Q-learning results in Self-correcting Q-learning. We show theoretically that this new algorithm enjoys the same convergence guarantees as Q-learning while being more accurate. Empirically, it performs better than Double Q-learning in domains with rewards of high variance, and it even attains faster convergence than Q-learning in domains with rewards of zero or low variance. These advantages transfer to a Deep Q Network implementation that we call Self-correcting DQN and which outperforms regular DQN and Double DQN on several tasks in the Atari 2600 domain.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191