Filters








13 Hits in 5.6 sec

E-PUR: An Energy-Efficient Processing Unit for Recurrent Neural Networks [article]

Franyell Silfa, Gem Dot, Jose-Maria Arnau, Antonio Gonzalez
2017 arXiv   pre-print
In this paper, we present E-PUR, an energy-efficient processing unit tailored to the requirements of LSTM computation.  ...  The main goal of E-PUR is to support large recurrent neural networks for low-power mobile devices.  ...  E-PUR PROCESSING UNIT In this section, we present E-PUR, an energy-efficient processing unit for large LSTM networks.  ... 
arXiv:1711.07480v1 fatcat:uii53e4uxbhr7cxgfg57elol3m

Neuron-Level Fuzzy Memoization in RNNs

Franyell Silfa, Gem Dot, Jose-Maria Arnau, Antonio Gonzàlez
2019 Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture - MICRO '52  
Recurrent Neural Networks (RNNs) are a key technology for applications such as automatic speech recognition or machine translation.  ...  For each application run, each recurrent layer is executed many times for processing a potentially large sequence of inputs (words, images, audio frames, etc.).  ...  We strive to select an energy-efficient configuration for all the neural networks in Table 1 .  ... 
doi:10.1145/3352460.3358309 dblp:conf/micro/SilfaDA019 fatcat:x3jsl53no5fadcdfml54goevem

Saving RNN Computations with a Neuron-Level Fuzzy Memoization Scheme [article]

Franyell Silfa, Jose-Maria Arnau, Antonio González
2022 arXiv   pre-print
Recurrent Neural Networks (RNNs) are a key technology for applications such as automatic speech recognition or machine translation.  ...  For each application run, recurrent layers are executed many times for processing a potentially large sequence of inputs (words, images, audio frames, etc.).  ...  We strive to select an energy-efficient configuration for all the neural networks in Table 1 .  ... 
arXiv:2202.06563v1 fatcat:nxrbufmshbemljguaaxnqbzf7a

E-BATCH: Energy-Efficient and High-Throughput RNN Batching [article]

Franyell Silfa, Jose Maria Arnau, Antonio Gonzalez
2020 arXiv   pre-print
In E-PUR, E-BATCH improves throughput by 1.8x and energy-efficiency by 3.6x, whereas in TPU, it improves throughput by 2.1x and energy-efficiency by 1.6x, over the state-of-the-art.  ...  Recurrent Neural Network (RNN) inference exhibits low hardware utilization due to the strict data dependencies across time-steps. Batching multiple requests can increase throughput.  ...  We first implement it on top of E-PUR, an energy-efficient accelerator for RNNs [3] . E-PUR consists of four computational units dedicated to evaluate the four gates in an LSTM cell.  ... 
arXiv:2009.10656v1 fatcat:26722qdnynehnawxyrpn6qbfpi

SHARP: An Adaptable, Energy-Efficient Accelerator for Recurrent Neural Network [article]

Reza Yazdani, Olatunji Ruwase, Minjia Zhang, Yuxiong He, Jose-Maria Arnau, Antonio Gonzalez
2022 arXiv   pre-print
The effectiveness of Recurrent Neural Networks (RNNs) for tasks such as Automatic Speech Recognition has fostered interest in RNN inference acceleration.  ...  efficiency for certain chosen model sizes.  ...  SHARP: An Adaptable, Energy-Efficient Accelerator for Recurrent Neural Network • 19  ... 
arXiv:1911.01258v2 fatcat:cljke2zczvgobds2txf66tdn44

Boosting LSTM Performance Through Dynamic Precision Selection [article]

Franyell Silfa, Jose-Maria Arnau, Antonio Gonzàlez
2019 arXiv   pre-print
The use of low numerical precision is a fundamental optimization included in modern accelerators for Deep Neural Networks (DNNs).  ...  The number of bits of the numerical representation is set to the minimum precision that is able to retain accuracy based on an offline profiling, and it is kept constant for DNN inference.  ...  Brainwave [21] is a Neural Processing Unit that achieves an order of magnitude improvement in latency and throughput over state-of-theart GPUs on large LSTM networks.  ... 
arXiv:1911.04244v1 fatcat:rrigdbhdvzcmlg45na2byg6aya

E-BATCH: Energy-Efficient and High-Throughput RNN Batching

Franyell Silfa, Jose Maria Arnau, Antonio González
2022 ACM Transactions on Architecture and Code Optimization (TACO)  
E-BATCH improves throughput by 1.8× and energy efficiency by 3.6× in E-PUR, whereas in TPU, it improves throughput by 2.1× and energy efficiency by 1.6×, over the state-of-the-art.  ...  Recurrent Neural Network (RNN) inference exhibits low hardware utilization due to the strict data dependencies across time-steps. Batching multiple requests can increase throughput.  ...  We first implement it on top of E-PUR, an energy-efficient accelerator for RNNs [29] . E-PUR consists of four computational units dedicated to evaluating the four gates in an LSTM cell.  ... 
doi:10.1145/3499757 fatcat:le6azhnu2zgovdglkw5tydrb2y

Boosting LSTM Performance Through Dynamic Precision Selection

Franyell Silfa, Jose Maria Arnau, Antonio Gonzalez
2020 2020 IEEE 27th International Conference on High Performance Computing, Data, and Analytics (HiPC)  
The use of low numerical precision is a fundamental optimization included in modern accelerators for Deep Neural Networks (DNNs).  ...  For a set of popular LSTM networks, it chooses the lowest precision for 57% of the time, outperforming systems that fix the precision statically.  ...  Recurrent Neural Networks A Recurrent Neural Network (RNN) is a state-of-theart machine learning algorithm that is very successful in sequence-to-sequence problems such as machine translation [1] and  ... 
doi:10.1109/hipc50609.2020.00046 fatcat:ganginwo5jcy5mbh372ohepgbq

MASR: A Modular Accelerator for Sparse RNNs [article]

Udit Gupta, Brandon Reagen, Lillian Pentecost, Marco Donato, Thierry Tambe, Alexander M. Rush, Gu-Yeon Wei, David Brooks
2019 arXiv   pre-print
Recurrent neural networks (RNNs) are becoming the de facto solution for speech recognition. RNNs exploit long-term temporal relationships in data by applying repeated, learned transformations.  ...  In comparison to current state-of-the-art sparse neural network accelerators (e.g., EIE), MASR provides 2x area 3x energy, and 1.6x performance benefits.  ...  We then simplify the neural networks to establish an efficient baseline RNN for ASR. A.  ... 
arXiv:1908.08976v1 fatcat:f2kbiwvdfzftfhu2rnhnqnguza

Challenging the Sense of Belonging Edited by

Maura Di, Mauro, Bettina Gehrke
unpublished
Culture and identity for this millennium are a non-linear dynamic, complex and ever evolving open process of integrated living systems of cognition and consciousness in the development of quality human  ...  making possible the integration process, and its value-added economic efficiency.  ...  Si tratta inevitabilmente di un passato selezionato e ristrutturato, ma è pur sempre la principale fonte di cui disponiamo per conoscere la nostra storia personale: è la nostra "scienza della realtà".  ... 
fatcat:qdowrft2ejbdre2kautodswyhe

Subtext and Context: Pathways to Trauma Recovery in Gianna Manzini's Ritratto in piedi

Maria Sergi
2020
Submitted for the degree of Doctor of Philosophy to the School of Humanities and Social Sciences, College of Arts, Social Sciences and Commerce, La Trobe University, Victoria, Australia.  ...  Recalling Manzini's description of the presence of wasps as threatening as machine-gun fire, the reader takes in the "strano e pur naturale concerto di suoni interrotti", (Manzini Ritratto 130), "strange  ...  very efficiently for at least seven hundred million years right up to the Pleistocene era when humanity entered an intense period of growth resulting in the need for major homeostatic adaptations to cope  ... 
doi:10.26181/5f9f8dbbef0a0 fatcat:q6ycq7cku5aclc743ajcsgr5eu

AIUCD 2022 - Proceedings

Fabio Ciracì, Giulia Miglietta, Carola Gatto
2022
ACKNOWLEDGEMENTS We are very grateful to Ondrej Uzovic for the numerous conversations about KRINO, argument-checking, and human annotations.  ...  BIBLIOGRAFIA ACKNOWLEDGEMENTS The RePIM in LOD project was founded by I Tatti -The Harvard University Center For Italian Renaissance Studies with a Digital Humanities Fellowship grant.  ...  This posed an issue for the stylometric analysis, as the attribution of short texts generally proves less efficient.  ... 
doi:10.6092/unibo/amsacta/6848 fatcat:fkiscqeu2vbpxlu737j3bacd2a

L'esperienza della Grande Guerra nell'autobiografia femminile in Gran Bretagna

Barbara Porzio
2009
He tightened his puttees and shook himself in his creased clothing, like a dog; and began to shout himself, for an orderly.  ...  Communication with the right broke down on an excited appeal for more bombs, and he sent off a scratch crowd, signalers, orderlies, bombers, telephonists, about twenty men,  ...  Data la natura così ambigua del testo, è lecito chiedersi perché la Brittain, pur modellando le esperienze della protagonista sulle proprie e pur avendo raggiunto il successo grazie ad un"autobiografia  ... 
doi:10.6092/unina/fedoa/3776 fatcat:gyr3souobve75hdqexjjlamb5a