Publications by Mario Paulo Drumond Lages De Oliveira

Equinox: Training (for Free) on a Custom Inference Accelerator

Babak Falsafi, Martin Jaggi, Ahmet Caner Yüzügüler, Louis Coulon, Mario Paulo Drumond Lages De Oliveira, Arash Pourhabibi Zarandi

DNN inference accelerators executing online services exhibit low average loads because of service demand variability, leading to poor resource utilization. Unfortunately, reclaiming idle inference cycles is difficult as other workloads can not execute on a ...

ACM2021

Data transformer apparatus

Babak Falsafi, Christoph Koch, Siddharth Gupta, Mario Paulo Drumond Lages De Oliveira, Mark Johnathon Sutherland, Arash Pourhabibi Zarandi, Zilu Tian, Hussein Kassir

Data transformer apparatus comprising at least a dispatcher module (D), a reader module (R), a converter module (C) and a writer module (W). The dispatcher module (D) is configured to: receive a data transformation request (DTR) including: a first informat ...

2021

Optimus Prime: Accelerating Data Transformation in Servers

Babak Falsafi, Christoph Koch, Siddharth Gupta, Mario Paulo Drumond Lages De Oliveira, Mark Johnathon Sutherland, Arash Pourhabibi Zarandi, Zilu Tian, Hussein Kassir

Modern online services are shifting away from monolithic applications to loosely-coupled microservices because of their improved scalability, reliability, programmability and development velocity. Microservices communicating over the datacenter network req ...

ACM2020

ColTraIn: Co-located DNN training and inference

Mario Paulo Drumond Lages De Oliveira

Deep neural network inference accelerators are deployed at scale to accommodate online services, but face low average load because of service demand variability, leading to poor resource utilization. Unfortunately, reclaiming inference idle cycles is diffi ...

EPFL2020

Analog Neural Networks with Deep-submicron Nonlinear Synapses

Pascal Frossard, Babak Falsafi, Ahmet Caner Yüzügüler, Mario Paulo Drumond Lages De Oliveira, Firat Çelik

Deep neural network (DNN) inference tasks are computationally expensive. Digital DNN accelerators offer better density and energy efficiency than general-purpose processors but still not sufficient to be deployable on resource-constrained settings.Analog c ...

2019

LTRF: Enabling High-Capacity Register Files for GPUs via Hardware/Software Cooperative Register Prefetching

Babak Falsafi, Mario Paulo Drumond Lages De Oliveira, Hamid Sarbazi-Azad, Seyed Borna Ehsani

Graphics Processing Units (GPUs) employ large register files to accommodate all active threads and accelerate context switching. Unfortunately, register files are a scalability bottleneck for future GPUs due to long access latency, high power consumption, ...

2018

Training DNNs with Hybrid Block Floating Point

Babak Falsafi, Martin Jaggi, Tao Lin, Mario Paulo Drumond Lages De Oliveira

The wide adoption of DNNs has given birth to unrelenting computing requirements, forcing datacenter operators to adopt domain-specific accelerators to train them. These accelerators typically employ densely packed full-precision floating-point arithmetic t ...

NEURAL INFORMATION PROCESSING SYSTEMS (NIPS)2018

The Mondrian Data Engine

Babak Falsafi, Boris Robert Grot, Alexandros Daglis, Javier Picorel Obando, Mario Paulo Drumond Lages De Oliveira, Dmitrii Ustiugov, Dionysios Pnevmatikatos, Nooshin Sadat Mirzadeh

The increasing demand for extracting value out of ever-growing data poses an ongoing challenge to system designers, a task only made trickier by the end of Dennard scaling. As the performance density of traditional CPU-centric architectures stagnates, adva ...

2017