An official website of the United States government
In this paper, we present a blockwise optimization method for masking-based networks (BLOOM-Net) for training scalable speech enhancement networks. Here, we design our network with a residual learning scheme and train the internal separator blocks sequentially to obtain a scalable masking-based deep neural network for speech enhancement. Its scalability lets it dynamically adjust the run-time complexity depending on the test time environment. To this end, we modularize our models in that they can flexibly accommodate varying needs for enhancement performance and constraints on the resources, incurring minimal memory or training overhead due to the added scalability. Our experiments on speech enhancement demonstrate that the proposed blockwise optimization method achieves the desired scalability with only a slight performance degradation compared to corresponding models trained end-to-end.
more »
« less
AdaLearner: An adaptive distributed mobile learning system for neural networks
Neural networks hold a critical domain in machine learning algorithms because of their self-adaptiveness and state-of-the-art performance. Before the testing (inference) phases in practical use, sophisticated training (learning) phases are required, calling for efficient training methods with higher accuracy and shorter converging time. Many existing studies focus on the training optimization on high-performance servers or computing clusters, e.g. GPU clusters. However, training neural networks on resource-constrained devices, e.g. mobile platforms, is an important research topic barely touched. In this paper, we implement AdaLearner–an adaptive distributed mobile learning system for neural networks that trains a single network with heterogenous mobile resources under the same local network in parallel. To exploit the potential of our system, we adapt neural networks training phase to mobile device-wise resources and fiercely decrease the transmission overhead for better system scalability. On three representative neural network structures trained from two image classification datasets, AdaLearner boosts the training phase significantly. For example, on LeNet, 1.75-3.37⇥ speedup is achieved when increasing the worker nodes from 2 to 8, thanks to the achieved high execution parallelism and excellent scalability.
more »
« less
- PAR ID:
- 10063491
- Date Published:
- Journal Name:
- IEEE/ACM International Conference on Computer-Aided Design (ICCAD)
- Page Range / eLocation ID:
- 291 to 296
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Deep neural networks (DNNs) have gained considerable attention in various real-world applications due to the strong performance on representation learning. However, a DNN needs to be trained many epochs for pursuing a higher inference accuracy, which requires storing sequential versions of DNNs and releasing the updated versions to users. As a result, large amounts of storage and network resources are required, which significantly hamper DNN utilization on resource-constrained platforms (e.g., IoT, mobile phone). In this paper, we present a novel delta compression framework called Delta-DNN, which can efficiently compress the float-point numbers in DNNs by exploiting the floats similarity existing in DNNs during training. Specifically, (1) we observe the high similarity of float-point numbers between the neighboring versions of a neural network in training; (2) inspired by delta compression technique, we only record the delta (i.e., the differences) between two neighboring versions, instead of storing the full new version for DNNs; (3) we use the error-bounded lossy compression to compress the delta data for a high compression ratio, where the error bound is strictly assessed by an acceptable loss of DNNs’ inference accuracy; (4) we evaluate Delta-DNN’s performance on two scenarios, including reducing the transmission of releasing DNNs over network and saving the storage space occupied by multiple versions of DNNs. According to experimental results on six popular DNNs, DeltaDNN achieves the compression ratio 2x~10x higher than state-ofthe-art methods, while without sacrificing inference accuracy and changing the neural network structure.more » « less
-
Open radio access networks (e.g., O-RAN) facilitate fine-grained control (e.g., near-RT RIC) in next-generation networks, necessitating advanced AI/ML techniques in handling online resource orchestration in real-time. However, existing approaches can hardly adapt to time-evolving network dynamics in network slicing, leading to significant online performance degradation. In this paper, we propose AdaSlicing, a new adaptive network slicing system, to online learn to orchestrate virtual resources while efficiently adapting to continual network dynamics. The AdaSlicing system includes a new soft-isolated RAN virtualization framework and a novel AdaOrch algorithm. We design the AdaOrch algorithm by integrating AI/ML techniques (i.e., Bayesian learning agents) and optimization methods (i.e., the ADMM coordinator). We design the soft-isolated RAN virtualization to improve the virtual resource utilization of slices while assuring the isolation among virtual resources at runtime. We implement AdaSlicing on an O-RAN compliant network testbed by using OpenAirInterface RAN, Open5GS Core, and FlexRIC near-RT RIC, with Ettus USRP B210 SDR. With extensive network experiments, we demonstrate that AdaSlicing substantially outperforms state-of-the-art works with 64.2% cost reduction and 45.5% normalized performance improvement, which verifies its high adaptability, scalability, and assurance.more » « less
-
ABSTRACT We present the results of a proof-of-concept experiment that demonstrates that deep learning can successfully be used for production-scale classification of compact star clusters detected in Hubble Space Telescope(HST) ultraviolet-optical imaging of nearby spiral galaxies ($$D\lesssim 20\, \textrm{Mpc}$$) in the Physics at High Angular Resolution in Nearby GalaxieS (PHANGS)–HST survey. Given the relatively small nature of existing, human-labelled star cluster samples, we transfer the knowledge of state-of-the-art neural network models for real-object recognition to classify star clusters candidates into four morphological classes. We perform a series of experiments to determine the dependence of classification performance on neural network architecture (ResNet18 and VGG19-BN), training data sets curated by either a single expert or three astronomers, and the size of the images used for training. We find that the overall classification accuracies are not significantly affected by these choices. The networks are used to classify star cluster candidates in the PHANGS–HST galaxy NGC 1559, which was not included in the training samples. The resulting prediction accuracies are 70 per cent, 40 per cent, 40–50 per cent, and 50–70 per cent for class 1, 2, 3 star clusters, and class 4 non-clusters, respectively. This performance is competitive with consistency achieved in previously published human and automated quantitative classification of star cluster candidate samples (70–80 per cent, 40–50 per cent, 40–50 per cent, and 60–70 per cent). The methods introduced herein lay the foundations to automate classification for star clusters at scale, and exhibit the need to prepare a standardized data set of human-labelled star cluster classifications, agreed upon by a full range of experts in the field, to further improve the performance of the networks introduced in this study.more » « less
-
Abstract Deep-learning models have become pervasive tools in science and engineering. However, their energy requirements now increasingly limit their scalability 1 . Deep-learning accelerators 2–9 aim to perform deep learning energy-efficiently, usually targeting the inference phase and often by exploiting physical substrates beyond conventional electronics. Approaches so far 10–22 have been unable to apply the backpropagation algorithm to train unconventional novel hardware in situ. The advantages of backpropagation have made it the de facto training method for large-scale neural networks, so this deficiency constitutes a major impediment. Here we introduce a hybrid in situ–in silico algorithm, called physics-aware training, that applies backpropagation to train controllable physical systems. Just as deep learning realizes computations with deep neural networks made from layers of mathematical functions, our approach allows us to train deep physical neural networks made from layers of controllable physical systems, even when the physical layers lack any mathematical isomorphism to conventional artificial neural network layers. To demonstrate the universality of our approach, we train diverse physical neural networks based on optics, mechanics and electronics to experimentally perform audio and image classification tasks. Physics-aware training combines the scalability of backpropagation with the automatic mitigation of imperfections and noise achievable with in situ algorithms. Physical neural networks have the potential to perform machine learning faster and more energy-efficiently than conventional electronic processors and, more broadly, can endow physical systems with automatically designed physical functionalities, for example, for robotics 23–26 , materials 27–29 and smart sensors 30–32 .more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICCAD.2017.8203791
