Discovering Multi-Hardware Mobile Models via Architecture Search

Chu, Grace; Arıkan, Okan; Bender, Gabriel; Brighton, Achille; Kindermans, Pieter-Jan; Liu, Hanxiao; Akin, Berkin; Gupta, Suyog

doi:10.1109/cvprw53098.2021.00337

Cited by 11 publications

(7 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…NAS and accelerator design. Hardware-aware NAS has been actively studied to incorporate characteristics of target device and automate the design of optimal architectures subject to latency and/or energy constraints [4,8,15,19,23,25,32,35]. These studies do not explore the hardware design space.…”

Section: Related Workmentioning

confidence: 99%

“…(DNN) models [36]. As DNNs are being deployed on increasingly diverse devices such as tiny Internet-of-Things devices, state-of-theart (SOTA) NAS is turning hardware-aware by further taking into consideration the target hardware as a crucial factor that affects the resulting performance (e.g., inference latency) of NAS-designed models [4,8,15,25,26,30,32] Likewise, optimizing hardware accelerators built on Field Programmable Gate Array (FPGA) or Application-Specific Integrated Circuit (ASIC), as well as the corresponding dataflows (e.g., scheduling DNN computations and mapping them on hardware), is also critical for speeding up DNN execution [1,11,33].…”

mentioning

confidence: 99%

See 1 more Smart Citation

A Semi-Decoupled Approach to Fast and Optimal Hardware-Software Co-Design of Neural Accelerators

Lu¹,

Yan²,

Shi³

et al. 2022

Preprint

View full text Add to dashboard Cite

In view of the performance limitations of fully-decoupled designs for neural architectures and accelerators, hardware-software codesign has been emerging to fully reap the benefits of flexible design spaces and optimize neural network performance. Nonetheless, such co-design also enlarges the total search space to practically infinity and presents substantial challenges. While the prior studies have been focusing on improving the search efficiency (e.g., via reinforcement learning), they commonly rely on co-searches over the entire architecture-accelerator design space. In this paper, we propose a semi-decoupled approach to reduce the size of the total design space by orders of magnitude, yet without losing optimality. We first perform neural architecture search to obtain a small set of optimal architectures for one accelerator candidate. Importantly, this is also the set of (close-to-)optimal architectures for other accelerator designs based on the property that neural architectures' ranking orders in terms of inference latency and energy consumption on different accelerator designs are highly similar. Then, instead of considering all the possible architectures, we optimize the accelerator design only in combination with this small set of architectures, thus significantly reducing the total search cost. We validate our approach by conducting experiments on various architecture spaces for accelerator designs with different dataflows. Our results highlight that we can obtain the optimal design by only navigating over the reduced search space. The source code of this work is at https://github.com/Ren-Research/CoDesign.

show abstract

Section: Related Workmentioning

confidence: 99%

mentioning

confidence: 99%

A Semi-Decoupled Approach to Fast and Optimal Hardware-Software Co-Design of Neural Accelerators

Lu¹,

Yan²,

Shi³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The pretraining dataset for CompressionNet is still Kinetics-600 [12]. We then reduced the model complexity by leveraging MoViNet [29] for 3D (CompressionNet) and MobileNet [30] To distinguish different versions of UVQ models, we refer to the initial model proposed in [25] as UVQ(ImageNet); the version using EfficientNet-b0 pretrained on JFT dataset, but otherwise the same as in [25], as UVQ (JFT); and the version using MoViNet and MobileNet as UVQ-lite (JFT). UVQ (ImageNet) and UVQ (JFT) have the same model complexity.…”

Section: Reducing Uvq Model Complexitymentioning

confidence: 99%

Revisiting the Efficiency of UGC Video Quality Assessment

Wang

Yim

Birkbeck

et al. 2022

2022 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

UGC video quality assessment (UGC-VQA) is a challenging research topic due to the high video diversity and limited public UGC quality datasets. State-of-the-art (SOTA) UGC quality models tend to use high complexity models, and rarely discuss the trade-off among complexity, accuracy, and generalizability. We propose a new perspective on UGC-VQA, and show that model complexity may not be critical to the performance, whereas a more diverse dataset is essential to train a better model. We illustrate this by using a light weight model, UVQ-lite, which has higher efficiency and better generalizability (less overfitting) than baseline SOTA models. We also propose a new way to analyze the sufficiency of the training set, by leveraging UVQ's comprehensive features. Our results motivate a new perspective about the future of UGC-VQA research, which we believe is headed toward more efficient models and more diverse datasets.

show abstract

“…In particular, we made modifications on top of the most recent version of MobileNet, MobileNet Multi-Hardware (MNMH) in [5], which is searched for mobile classification with AutoML considering applications on various chips. The specifications are shown in Table 1.…”

Section: Tailored Mobilenet Backbonementioning

confidence: 99%

“…One common practice is to adapt a strong backbone network built for image classification task to the segmentation task. To maintain the highresolution feature representation for this dense prediction task, excessive down-sampling or striding operations at the Latency is simply normalized using real on-device latency on various platforms as [5]. Multiple MOSAIC models with different filter sizes are shown on the Cityscapes plot.…”

Section: Introductionmentioning

confidence: 99%

MOSAIC: Mobile Segmentation via decoding Aggregated Information and encoded Context

Howard¹

2021

Preprint

Self Cite

View full text Add to dashboard Cite

We present a next-generation neural network architecture, MOSAIC, for efficient and accurate semantic image segmentation on mobile devices. MOSAIC is designed using commonly supported neural operations by diverse mobile hardware platforms for flexible deployment across various mobile platforms. With a simple asymmetric encoderdecoder structure which consists of an efficient multi-scale context encoder and a light-weight hybrid decoder to recover spatial details from aggregated information, MOSAIC achieves new state-of-the-art performance while balancing accuracy and computational cost. Deployed on top of a tailored feature extraction backbone based on a searched classification network, MOSAIC achieves a 5% absolute accuracy gain surpassing the current industry standard MLPerf models and state-of-the-art architectures.

show abstract

Discovering Multi-Hardware Mobile Models via Architecture Search

Cited by 11 publications

References 5 publications

A Semi-Decoupled Approach to Fast and Optimal Hardware-Software Co-Design of Neural Accelerators

A Semi-Decoupled Approach to Fast and Optimal Hardware-Software Co-Design of Neural Accelerators

Revisiting the Efficiency of UGC Video Quality Assessment

MOSAIC: Mobile Segmentation via decoding Aggregated Information and encoded Context

Contact Info

Product

Resources

About