Prototype-Based Personalized Pruning

@article{Kim2021PrototypeBasedPP,
  title={Prototype-Based Personalized Pruning},
  author={Jang-Hyun Kim and Simyung Chang and Sungrack Yun and Nojun Kwak},
  journal={ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year={2021},
  pages={3925-3929}
}
  • Jang-Hyun Kim, Simyung Chang, Nojun Kwak
  • Published 25 March 2021
  • Computer Science
  • ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Nowadays, as edge devices such as smartphones become prevalent, there are increasing demands for personalized services. However, traditional personalization methods are not suitable for edge devices because retraining or finetuning is needed with limited personal data. Also, a full model might be too heavy for edge devices with limited resources. Unfortunately, model compression methods which can handle the model complexity issue also require the retraining phase. These multiple training phases… 

Figures and Tables from this paper

PQK: Model Compression via Pruning, Quantization, and Knowledge Distillation
TLDR
A novel model compression method for the devices with limited computational resources, called PQK consisting of pruning, quantization, and knowledge distillation processes, which makes use of unimportant weights pruned in the pruning process to make a teacher network for training a better student network without pre-training the teacher model.
A Personalized Benchmark for Face Anti-spoofing
TLDR
This paper presents a complementary approach to augment existing face anti-spoofing benchmarks to account for enrollment images associated with each query image, and applies this strategy on two recently introduced datasets: CelebA-Spoof [53] and SiW [29].
PrimeNet: Adaptive Multi-Layer Deep Neural Structure for Enhanced Feature Selection in Early Convolution Stage
TLDR
The PrimeNet unit is proposed as an alternative to traditional convolution and dense blocks for faster and memory-efficient training, outperforming previously reported results aimed at adaptive methods for parallel and multilayer deep neural systems.
PrimeNet: Adaptive Multi-Layer Deep Neural Structure for Enhanced Feature Selection in Early Convolution Stage
TLDR
The PrimeNet unit is proposed as an alternative to traditional convolution and dense blocks for faster and memory-efficient training, outperforming previously reported results aimed at adaptive methods for parallel and multilayer deep neural systems.

References

SHOWING 1-10 OF 22 REFERENCES
URNet : User-Resizable Residual Networks with Conditional Gating Module
TLDR
This work proposes User-Resizable Residual Networks (URNet), which allows users to adjust the computational cost of the network as needed during evaluation, and maintains the accuracy of the baseline even when resizing it to approximately 80% of the original network.
Runtime Neural Pruning
TLDR
A Runtime Neural Pruning (RNP) framework which prunes the deep neural network dynamically at the runtime and preserves the full ability of the original network and conducts pruning according to the input image and current feature maps adaptively.
Position-based Scaled Gradient for Model Quantization and Sparse Training
TLDR
The proposed position-based scaled gradient (PSG) reduces the gap between the weight distributions of a full-precision model and its compressed counterpart and enables the versatile deployment of a model either as an uncompressed mode or as a compressed mode depending on the availability of resources.
Deep Residual Learning for Small-Footprint Keyword Spotting
  • Raphael Tang, Jimmy J. Lin
  • Computer Science, Economics
    2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
  • 2018
TLDR
This work explores the application of deep residual learning and dilated convolutions to the keyword spotting task, using the recently-released Google Speech Commands Dataset as a benchmark and establishes an open-source state-of-the-art reference to support the development of future speech-based interfaces.
Query-by-Example On-Device Keyword Spotting
TLDR
A threshold prediction method while using the user-specific keyword hypothesis only is proposed, which generates query-specific negatives by rearranging each query utterance in waveform and decides the threshold based on the enrollment queries and generated negatives.
On-Device Few-Shot Personalization for Real-Time Gaze Estimation
TLDR
The proposed supervised method achieves better accuracy using as few as 2-5 calibration points per user compared to prior methods that require more than 13 calibration points, and an unsupervised personalization method which uses only unlabeled facial images to improve gaze estimation accuracy.
The Technology Behind Personal Digital Assistants: An overview of the system architecture and key components
  • R. Sarikaya
  • Computer Science
    IEEE Signal Processing Magazine
  • 2017
TLDR
An overview of personal digital assistants (PDAs) is given; the system architecture, key components, and technology behind them are described; and their future potential to fully redefine human?computer interaction is discussed.
Convolutional Networks with Adaptive Inference Graphs
TLDR
This work proposes convolutional networks with adaptive inference graphs (ConvNet-AIG) that adaptively define their network topology conditioned on the input image, and observes that ConvNet- AIG shows a higher robustness than ResNets, complementing other known defense mechanisms.
Deep Compression: Compressing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding
TLDR
This work introduces "deep compression", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy.
Prototypical Networks for Few-shot Learning
TLDR
This work proposes Prototypical Networks for few-shot classification, and provides an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning.
...
...