11,966 Hits in 1.8 sec

A Programmable Approach to Model Compression [article]

Vinu Joseph, Saurav Muralidharan, Animesh Garg, Michael Garland, Ganesh Gopalakrishnan
2019 arXiv   pre-print
Deep neural networks frequently contain far more weights, represented at a higher precision, than are required for the specific task which they are trained to perform.  ...  In this paper, we introduce a programmable system for model compression called Condensa. Users programmatically compose simple operators, in Python, to build complex compression strategies.  ...  This programmable approach to model compression enables users to experiment and rapidly converge to an ideal scheme for a given compression context, avoiding manual trial-and-error search.  ... 
arXiv:1911.02497v1 fatcat:mrd3phetvjgptbtz4gjg77dmcm

Applying Data Compression Techniques on Systolic Neural Network Accelerator [article]

Navid Mirnouri
2016 arXiv   pre-print
Another technique is Data compression which is used in memory systems in order to save capacity and bandwidth.  ...  Acceleration is another solution that uses specialized logics in order to do computations in a way that is more power efficient.  ...  Implementation of neural networks can be both in hardware [3] [12] [13] or software [11] [14] . 3.1 Systolic Neural Network accelerator in Programmable logics [3] SNNAP is a neural network accelerator  ... 
arXiv:1701.03734v1 fatcat:nc7alulx3bbtxczw62wfekumie

Machine Learning for Systems

Heiner Litz, Milad Hashemi
2020 IEEE Micro  
The second article, "A Programmable Approach to Neural Network Compression," explores compression and pruning techniques to reduce the size and computational complexity of deep neural networks.  ...  In "RELEQ: A Reinforcement Learning Approach for Automatic Deep Quantization of Neural Networks," the authors describe a reinforcement learning mechanism to optimize deep neural network architectures.  ... 
doi:10.1109/mm.2020.3016551 fatcat:7lbbknjtmjamha2ek5fnc2tbem

Page 422 of SMPTE Motion Imaging Journal Vol. 103, Issue 7 [page]

1994 SMPTE Motion Imaging Journal  
Zortea The use of a neural network as an alternate architecture for video camera processing is investigated, and a novel approach to training and reducing the size of the network is presented.  ...  Programmability, then, indicates that a neural network may be useful for this application.  ... 

Optimising Hardware Accelerated Neural Networks with Quantisation and a Knowledge Distillation Evolutionary Algorithm

Robert Stewart, Andrew Nowlan, Pascal Bacchus, Quentin Ducasse, Ekaterina Komendantskaya
2021 Electronics  
Parallelising FPGA implementations of 2 and 3 bit quantised neural networks increases throughput from 6 k FPS to 373 k FPS, a 62× speedup.  ...  This paper compares the latency, accuracy, training time and hardware costs of neural networks compressed with our new multi-objective evolutionary algorithm called NEMOKD, and with quantisation.  ...  There are other compression methods such as weight sharing [48] to consider for hybrid compression. A complete study of neural network compression approaches is in [21] .  ... 
doi:10.3390/electronics10040396 fatcat:nxaxzcnx75fffp74vrve3ywnci

Digital electronics in fibres enable fabric-based machine-learning inference

Gabriel Loke, Tural Khudiyev, Brian Wang, Stephanie Fu, Syamantak Payra, Yorai Shaoul, Johnny Fung, Ioannis Chatziveroglou, Pin-Wen Chou, Itamar Chinn, Wei Yan, Anna Gitelson-Kahn (+2 others)
2021 Nature Communications  
The ability to realise digital devices within a fibre strand which can not only measure and store physiological parameters, but also harbour the neural networks required to infer sensory data, presents  ...  Here, a scalable preform-to-fibre approach is used to produce tens of metres of flexible fibre containing hundreds of interspersed, digital temperature sensors and memory devices with a memory density  ...  After training, the values of the weights and biases are extracted and reduced to two significant figures to produce a compressed neural network.  ... 
doi:10.1038/s41467-021-23628-5 pmid:34083521 fatcat:b564yhz5tfcibbg2rtlkn73f7m

Full Resolution Image Compression with Deep Neural Networks

Sarabjot Singh Grewal
2017 International Journal for Research in Applied Science and Engineering Technology  
In this research, we presented image compression and error correction control using deep neural networks. First, Deep Neural Network (DNN) is implemented for image compression.  ...  It reduces both time and bandwidth cost and leads to fast and efficient sharing of images and other information. Neural networks have been researching rapidly for image processing.  ...  K. (2012) presented an approach to a neural network for the security, authentication, and compression of the image.  ... 
doi:10.22214/ijraset.2017.9163 fatcat:4tki6tet6ja3pib36q74hsz5sm

Intelligent Meta-Imagers: From Compressed to Learned Sensing [article]

Chloé Saigre-Tardif, Rashid Faqiri, Hanting Zhao, Lianlin Li, Philipp del Hougne
2022 arXiv   pre-print
Computational meta-imagers synergize metamaterial hardware with advanced signal processing approaches such as compressed sensing.  ...  Here, we comprehensively review the evolution of computational meta-imaging from the earliest frequency-diverse compressive systems to modern programmable intelligent meta-imagers.  ...  (neural network nodes).  ... 
arXiv:2110.14022v3 fatcat:guzpgc257fg5pl7wdvv46u6b64

Best Papers From Hot Chips 32

Priyanka Raina, Cliff Young
2021 IEEE Micro  
"Compute Substrate for Software 2.0" explains startup TensTorrent's unique architecture for neural network acceleration, which takes a packet-network-inspired approach to control and flexibility, allowing  ...  better support for sparsity, varying numerical precision, and compression than older, more monolithic neural network accelerators.  ... 
doi:10.1109/mm.2021.3060294 fatcat:u3ajuboc3navlpy5mh5i6smxoy

Use of Artificial Neural Network in Design of Fly Ash Blended Cement Concrete Mixes

2019 International journal of recent technology and engineering  
Use of artificial neural networks (ANNs) for the checking of design composition of fly ash blended cement concrete mixes which were designed as per Indian standard guidelines has been made.  ...  Prediction of strength of such mixes at a later date by ANN has also been explored in this study.  ...  PROGRAMME OF THE STUDY In this study a simple back propagation neural network has been used to model the design mix proportions for the fly ash blended concrete mixes.  ... 
doi:10.35940/ijrte.c5146.098319 fatcat:ao3erwqxavb3ziskj2rm632wz4

Reduced memory region based deep Convolutional Neural Network detection

Denis Tome, Luca Bondi, Luca Baroffio, Stefano Tubaro, Emanuele Plebani, Danilo Pau
2016 2016 IEEE 6th International Conference on Consumer Electronics - Berlin (ICCE-Berlin)  
This paper makes two main contributions: (1) it proves that a region based deep neural network can be finely tuned to achieve adequate accuracy for pedestrian detection (2) it achieves a very low memory  ...  In order to achieve very high accuracy, recent pedestrian detectors have been based on Convolutional Neural Networks (CNN).  ...  In section II a review of the state of the art about neural network compression is offered to the reader.  ... 
doi:10.1109/icce-berlin.2016.7684706 dblp:conf/icce-berlin/TomeBBTPP16 fatcat:6u7eb6af6vdhtndztdby3qhqei

A Machine Learning-Assisted Numerical Predictor for Compressive Strength of Geopolymer Concrete Based on Experimental Data and Sensitivity Analysis

An Thao Huynh, Quang Dang Nguyen, Qui Lieu Xuan, Bryan Magee, TaeChoong Chung, Kiet Tuan Tran, Khoa Tan Nguyen
2020 Applied Sciences  
Methods assessed included artificial neural network (ANN), deep neural network (DNN), and deep residual network (ResNet), based on experimentally collected data.  ...  Geopolymer concrete offers a favourable alternative to conventional Portland concrete due to its reduced embodied carbon dioxide (CO2) content.  ...  In addition, the proposed machine learning approaches adopted a general training scheme for neural networks with standard input and output features, indicating promising potential to be applied to other  ... 
doi:10.3390/app10217726 fatcat:n37zeuutxbdzlbonkxkpxmwpti

SOCC 2019 Author Index

2019 2019 32nd IEEE International System-on-Chip Conference (SOCC)  
Workload Allocation for Distributed Deep Neural Networks in Edge-Cloud Continuum Zheng, Yu W1C.1 45 A Lossless Astronomical Data Compression Scheme with FPGA Acceleration Zhou, Jun W3C.3 181  ...  for Hardware Efficient Compressed Sensing Encoder Design Zhen, Degen W3C.1 171 A 4K Vision Computing Platform with Convolutional Neural Network Engine on FPGA Zheng, Lirong WP.5 213 Energy-Aware  ... 
doi:10.1109/socc46988.2019.9088078 fatcat:wzmgb42mjzbu5hvgccttlhw3zu

Detecting Compressed Cleartext Traffic from Consumer Internet of Things Devices [article]

Daniel Hahn, Noah Apthorpe, Nick Feamster
2018 arXiv   pre-print
We apply three machine learning models and achieve a maximum 66.9% accuracy with a convolutional neural network trained on raw packet data.  ...  Here, we present the first technique to automatically distinguish encrypted from compressed unencrypted network transmissions on a per-packet basis.  ...  ACKNOWLEDGMENTS Thanks to the developers of SciKit Learn, numpy, dpkt, Keras, TensorFlow, Python, AES, Blowfish, Gzip, Bzip2, rar, 7zip, and RawCap for making their software freely available.  ... 
arXiv:1805.02722v1 fatcat:thbtz6atuvhdlm3kdrexaqreo4

MASPINN: novel concepts for a neuroaccelerator for spiking neural networks

T. Schoenauer, N. Mehrtash, Andreas Jahnke, H. Klar, Thomas Lindblad, Mary Lou Padgett, Jason M. Kinser
1999 Ninth Workshop on Virtual Intelligence/Dynamic Neural Networks  
The accelerator architecture exploits two novel concepts for an efficient computation of spiking neural networks: weight caching and a compressed memory organization.  ...  We present the basic architecture of a Memory Optimized Accelerator for Spiking Neural Networks (MASPINN).  ...  IP-Memory (compressed)  ... 
doi:10.1117/12.343072 fatcat:ghez3fyovjhdvkqgywgvm2k7ly
« Previous Showing results 1 — 15 out of 11,966 results