An official website of the United States government
AI-powered applications often involve multiple deep neural network (DNN)-based prediction tasks to support application level functionalities. However, executing multi-DNNs can be challenging due to the high resource demands and computation costs that increase linearly with the number of DNNs. Multi-task learning (MTL) addresses this problem by designing a multi-task model that shares parameters across tasks based on a single backbone DNN. This paper explores an alternative approach called model fusion: rather than training a single multi-task model from scratch as MTL does, model fusion fuses multiple task-specific DNNs that are pre-trained separately and can have heterogeneous architectures into a single multi-task model. We materialize model fusion in a software framework called GMorph to accelerate multi- DNN inference while maintaining task accuracy. GMorph features three main technical contributions: graph mutations to fuse multi-DNNs into resource-efficient multi-task models, search-space sampling algorithms, and predictive filtering to reduce the high search costs. Our experiments show that GMorph can outperform MTL baselines and reduce the inference latency of multi-DNNs by 1.1-3X while meeting the target task accuracy.
more »
« less
SQUEEZENET: ALEXNET-LEVEL ACCURACY WITH 50X FEWER PARAMETERS AND <0.5MB MODEL SIZE
Recent research on deep neural networks has focused primarily on improving accuracy. For a given accuracy level, it is typically possible to identify multiple DNN architectures that achieve that accuracy level. With equivalent accuracy, smaller DNN architectures offer at least three advantages: (1) Smaller DNNs require less communication across servers during distributed training. (2) Smaller DNNs require less bandwidth to export a new model from the cloud to an autonomous car. (3) Smaller DNNs are more feasible to deploy on FPGAs and other hardware with limited memory. To provide all of these advantages, we propose a small DNN architecture called SqueezeNet. SqueezeNet achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. Additionally, with model compression techniques we are able to compress SqueezeNet to less than 0.5MB (510x smaller than AlexNet).
more »
« less
- Award ID(s):
- 1648576
- PAR ID:
- 10026154
- Date Published:
- Journal Name:
- arXiv.org
- ISSN:
- 2331-8422
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
AI-powered applications often involve multiple deep neural network (DNN)-based prediction tasks to support application level functionalities. However, executing multi-DNNs can be challenging due to the high resource demands and computation costs that increase linearly with the number of DNNs. Multi-task learning (MTL) addresses this problem by designing a multi-task model that shares parameters across tasks based on a single backbone DNN. This paper explores an alternative approach called model fusion: rather than training a single multi-task model from scratch as MTL does, model fusion fuses multiple task-specific DNNs that are pre-trained separately and can have heterogeneous architectures into a single multi-task model. We materialize model fusion in a software framework called GMorph to accelerate multi- DNN inference while maintaining task accuracy. GMorph features three main technical contributions: graph mutations to fuse multi-DNNs into resource-efficient multi-task models, search-space sampling algorithms, and predictive filtering to reduce the high search costs. Our experiments show that GMorph can outperform MTL baselines and reduce the inference latency of multi-DNNs by 1.1-3X while meeting the target task accuracy.more » « less
-
Variability-induced accuracy degradation of RRAM based DNNs is of great concern due to their significant potential for use in future energy-efficient machine learning architectures. To address this, we propose a two-step process. First, an enhanced testing procedure is used to predict DNN accuracy from a set of compact test stimuli (images). This test response (signature) is simply the concatenated vectors of output neurons of intermediate final DNN layers over the compact test images applied. DNNs with a predicted accuracy below a threshold are then tuned based on this signature vector. Using a clustering based approach, the signature is mapped to the optimal tuning parameter values of the DNN (determined using off-line training of the DNN via backpropagation) in a single step, eliminating any post-manufacture training of the DNN weights (expensive). The tuning parameters themselves consist of the gains and offsets of the ReLU activation of neurons of the DNN on a per-layer basis and can be tuned digitally. Tuning is achieved in less than a second of tuning time, with yield improvements of over 45% with a modest accuracy reduction of 4% compared to digital DNNs.more » « less
-
Variability-induced accuracy degradation of RRAMbased DNNs is of great concern due to their significant potential for use in future energy-efficient machine learning architectures. To address this, we propose a two-step process. First, an enhanced testing procedure is used to predict DNN accuracy from a set of compact test stimuli (images). This test response (signature) is simply the concatenated vectors of output neurons of intermediate and final DNN layers over the compact test images applied. DNNs with a predicted accuracy below a threshold are then tuned based on this signature vector. Using a clustering based approach, the signature is mapped to the optimal tuning parameter values of the DNN (determined using off-line training of the DNN via backpropagation) in a single step, eliminating any post-manufacture training of the DNN weights (expensive). The tuning parameters themselves consist of the gains and offsets of the ReLU activation of neurons of the DNN on a per-layer basis and can be tuned digitally. Tuning is achieved in less than a second of tuning time, with yield improvements of over 45% with a modest accuracy reduction of 4% compared to digital DNNs.more » « less
-
Agaian, Sos S.; DelMarco, Stephen P.; Asari, Vijayan K. (Ed.)Iris recognition is a widely used biometric technology that has high accuracy and reliability in well-controlled environments. However, the recognition accuracy can significantly degrade in non-ideal scenarios, such as off-angle iris images. To address these challenges, deep learning frameworks have been proposed to identify subjects through their off-angle iris images. Traditional CNN-based iris recognition systems train a single deep network using multiple off-angle iris image of the same subject to extract the gaze invariant features and test incoming off-angle images with this single network to classify it into same subject class. In another approach, multiple shallow networks are trained for each gaze angle that will be the experts for specific gaze angles. When testing an off-angle iris image, we first estimate the gaze angle and feed the probe image to its corresponding network for recognition. In this paper, we present an analysis of the performance of both single and multimodal deep learning frameworks to identify subjects through their off-angle iris images. Specifically, we compare the performance of a single AlexNet with multiple SqueezeNet models. SqueezeNet is a variation of the AlexNet that uses 50x fewer parameters and is optimized for devices with limited computational resources. Multi-model approach using multiple shallow networks, where each network is an expert for a specific gaze angle. Our experiments are conducted on an off-angle iris dataset consisting of 100 subjects captured at 10-degree intervals between -50 to +50 degrees. The results indicate that angles that are more distant from the trained angles have lower model accuracy than the angles that are closer to the trained gaze angle. Our findings suggest that the use of SqueezeNet, which requires fewer parameters than AlexNet, can enable iris recognition on devices with limited computational resources while maintaining accuracy. Overall, the results of this study can contribute to the development of more robust iris recognition systems that can perform well in non-ideal scenarios.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- The DOI is not currently available.
