About the Author:
"Accelerating Sparse Linear Algebra at NVIDIA
Senior Software Engineer, IC5
I’m a Software Engineer at NVIDIA in Santa Clara, US (headquarter). I work in the CUDA HPC Math Library team and have been leading the Sparse Linear Algebra group for the past 6 years.
My interests outside sparse linear algebra include parallel computing, software optimization, and graph algorithms. I am also passionate about programming and software engineering, primarily in C++.
.....
I taught C++ professionally for 5 years in undergraduate and graduate courses. I maintain an open-access C++ course. "
NISAR is the first of its kind mission, jointly developed by ISRO and NASA. It is an L and S-band, global, microwave imaging mission, with capability to acquire fully polarimetric and interferometric data.
The unique dual-band Synthetic Aperture Radar of NISAR employs advanced, novel SweepSAR technique, which provides high resolution and large swath imagery. NISAR will image the global land and ice-covered surfaces, including islands, sea-ice and selected oceans every 12 days.
NISAR mission’s primary objectives are to study land & ice deformation, land ecosystems, and oceanic regions in areas of common interest to the US and Indian science communities.
NISAR mission will help to
measure the woody biomass and its changes
track changes in the extent of active crops
understand the changes in wetlands’ extent
map Greenland’s & Antarctica’s ice sheets, dynamics of sea ice and mountain glaciers
characterize land surface deformation related to seismicity, volcanism, landslides, and subsidence & uplift associated with changes in subsurface aquifers, hydrocarbon reservoirs, etc.
Spacecraft Configuration
The Spacecraft is built around ISRO’s I-3K Structure. It carries two major Payloads viz., L & S- Band Synthetic Aperture Radar (SAR).
The S-band Radar system, data handling & high- speed downlink system, the spacecraft and the launch system are developed by ISRO. The L-band Radar system, high speed downlink system, the Solid-State Recorder, GPS receiver, the 9m Boom hoisting the 12m reflector are delivered by NASA.
Further, ISRO takes care of the satellite commanding and operations, NASA will provide the orbit maneuver plan and RADAR operations plan.
NISAR mission will be aided with ground station support of both ISRO and NASA for downloading of the acquired images, which after the necessary processing will be disseminated to the user community
The data acquired through S-band and L-band SAR from a single platform will help the scientists to understand the changes happening to Planet Earth."
"‘You’re a traitor!’ yelled the boy. ‘You’re a thought criminal! You’re a Eurasian spy! I’ll shoot you, I’ll vaporize you, I’ll send you to the salt mines!’
Suddenly they were both leaping around him, shouting ‘Traitor!’ and ‘Thought-criminal!’, the little girl imitating her brother in every movement.
It was somehow slightly frightening, like the gamboling of tiger cubs which will soon grow up into man-eaters."
Many poor countries are perfectly capable of manufacturing their own medical equipment
Background: I have worked in hospitals in Bihar & Vellore Tamilnadu, and have university mates, friends and family who worked in hospitals in Mizoram, PNG, Libya, Nepal, Uganda and even South Sudan.
Even the doctors who worked in South Sudan had no difficulty sourcing basic equipment. They did however, have enormous trouble sourcing funding and trained personnel who were willing to work in a potentially lethal situation.
"This note clarifies the relationship between the recent TurboQuant work and the earlier DRIVE (NeurIPS 2021) and EDEN (ICML 2022) schemes. DRIVE is a 1-bit quantizer that EDEN extended to any bits per coordinate; we refer to them collectively as EDEN.
First, TurboQuant is a special case of EDEN obtained by fixing EDEN's scalar scale parameter to . EDEN supports both biased and unbiased quantization, each optimized by a different (chosen via methods described in the EDEN works). The fixed choice used by TurboQuant is generally suboptimal, although the optimal for biased EDEN converges to as the dimension grows; accordingly TurboQuant approaches EDEN's behavior for large .
Second, TurboQuant combines a biased -bit EDEN step with an unbiased 1-bit QJL quantization of the residual. It is suboptimal in three ways: (1) its -bit step uses the suboptimal ; (2) its 1-bit unbiased residual quantization has worse MSE than (unbiased) 1-bit EDEN; (3) chaining a biased -bit step with a 1-bit unbiased residual step is inferior to unbiasedly quantizing the input directly with -bit EDEN.
Third, some of the analysis in the TurboQuant work mirrors that of the EDEN works: both exploit the connection between random rotations and the shifted Beta distribution, use the Lloyd-Max algorithm, and note that Randomized Hadamard Transforms can replace uniform random rotations.
Experiments support these claims: biased EDEN (with optimized ) is more accurate than TurboQuant, and unbiased EDEN is markedly more accurate than TurboQuant, often by more than a bit (e.g., 2-bit EDEN beats 3-bit TurboQuant). We also repeat all accuracy experiments from the TurboQuant paper, showing that EDEN outperforms it in every setup we have tried."
". While it represents one bit of information, it is typically implemented as 1 byte in arrays, and often 4 bytes (an int) or more as a standalone variable on the stack "
I’m a Software Engineer at NVIDIA in Santa Clara, US (headquarter). I work in the CUDA HPC Math Library team and have been leading the Sparse Linear Algebra group for the past 6 years.
My interests outside sparse linear algebra include parallel computing, software optimization, and graph algorithms. I am also passionate about programming and software engineering, primarily in C++.
.....
I taught C++ professionally for 5 years in undergraduate and graduate courses. I maintain an open-access C++ course. "
https://federico-busato.github.io/
reply