Oleg Zabluda's blog: November 27, 2018

Oleg Zabluda's blog

Tuesday, November 27, 2018

Not our school district
Not our school district
https://www.mercurynews.com/2018/11/14/horgan-grim-pupil-numbers-are-forcing-the-issue-in-redwood-city/

Labels: Oleg Zabluda

"""
"""
The authors proposed a model (BigGAN) with modifications focused on the following three aspects:

- Scalability: As the authors discovered that GANs benefit dramatically from scaling, they introduced two architectural changes to improve scalability (described in detail in the paper’s Appendix B), while at the same time improving conditioning by applying orthogonal regularization to the generator.

- Robustness: The orthogonal regularization applied to the generator makes the model amenable to the “truncation trick” so that fine control of the trade-offs between fidelity and variety is possible by truncating the latent space.

- Stability: The authors discovered and characterized instabilities specific to large-scale GANs, and devised solutions to minimize the instabilities — although these involved a relatively high trade-off on performance.

In addition to its performance boost at 128×128 resolutions, BigGAN also outperformed the previous SotA at 256×256 and 512×512 resolutions on ImageNet.
"""
https://medium.com/syncedreview/biggan-a-new-state-of-the-art-in-image-synthesis-cf2ec5694024
https://medium.com/syncedreview/biggan-a-new-state-of-the-art-in-image-synthesis-cf2ec5694024

Labels: Oleg Zabluda

Simple, Distributed, and Accelerated Probabilistic Programming (2018)
Simple, Distributed, and Accelerated Probabilistic Programming (2018)
Dustin Tran, Matthew Hoffman, Dave Moore, Christopher Suter, Srinivas Vasudevan, Alexey Radul, Matthew Johnson, Rif A. Saurous
"""
We describe a simple, low-level approach for embedding probabilistic programming in a deep learning ecosystem. In particular, we distill probabilistic programming down to a single abstraction---the random variable. Our lightweight implementation in TensorFlow enables numerous applications: a model-parallel variational auto-encoder (VAE) with 2nd-generation tensor processing units (TPUv2s); a data-parallel autoregressive model (Image Transformer) with TPUv2s; and multi-GPU No-U-Turn Sampler (NUTS). For both a state-of-the-art VAE on 64x64 ImageNet and Image Transformer on 256x256 CelebA-HQ, our approach achieves an optimal linear speedup from 1 to 256 TPUv2 chips. With NUTS, we see a 100x speedup on GPUs over Stan and 37x over PyMC3.
"""
https://arxiv.org/abs/1811.02091
https://arxiv.org/abs/1811.02091

Labels: Oleg Zabluda

"""
"""
Facebook AI Research (FAIR) and NYU School of Medicine’s Center for Advanced Imaging Innovation and Research (CAI²R) are sharing new open source tools and data as part of fastMRI, a joint research project to spur development of AI systems to speed MRI scans by up to 10x. Today’s releases include new [U-net] AI models and baselines for this task. It also includes the first large-scale MRI data set of its kind, which can serve as a benchmark for future research. [...] 1.5 million MR images drawn from 10,000 scans, as well as raw measurement data from nearly 1,600 scans.

By sharing a standardized set of AI tools and MRI data, as well as hosting a leaderboard where research teams can compare their results, we aim to help improve diagnostic imaging technology, and eventually increase patients’ access to a powerful and sometimes life-saving technology. With new AI techniques, we hope to generate scans that require much less measurement data to produce the image detail necessary for accurate detection of abnormalities.
[...]
In the more than four decades since medical MR imaging was introduced, researchers have tried consistently to shorten the technology’s long scan times, which can sometimes require patients to remain stationary for more than an hour. [...] MRI devices collect a series of individual 2D spatial measurements — known as k-space data in the medical imaging community — and convert them into various images. By training neural networks on a large amounts of k-space data, this image reconstruction technique allows for less detailed initial scans,
[...]
The ultimate goal of the fastMRI project is to use AI-driven image reconstruction to achieve up to a 10x reduction in scan times. To begin, we’re providing baseline models for ML-based image reconstruction from k-space data subsampled at 4x and 8x scan accelerations. And we’ve already seen promising preliminary results for accelerating MR imaging by up to four times.
[...]
The raw measurement data in this data set sets it apart from previous MR databases and could prove particularly valuable for researchers. It consists of the k-space data that’s collected during scanning and typically discarded after it’s used to generate images. [...] The fastMRI data set also includes undersampled versions of those measurements, with k-space lines retrospectively masked, to simulate partial-data scans.

The k-space data in this data set is drawn from MR devices with multiple magnetic coils, and it also includes data that simulates measurements from single-coil machines. Though research related to accelerating multi-coil scans is more relevant for clinical practice — they’re more precise and more common than single-coil scans — the inclusion of single-coil k-space data measurements offers AI researchers an entry point for applying ML to this imaging task. [...] we focused on two tasks — single-coil reconstruction and multi-coil reconstruction. [...] In both the single-coil and multi-coil deep learning baselines, our models are based on u-nets [...] Our initial work has indicated that the u-net architecture is particularly responsive to training on a large amount of data, such as the MR reconstruction data set released by NYU School of Medicine. [...] NYU School of Medicine plans to release additional images and measurements in the near future, related to brain and liver scans (the current data set consists of knee scans).
"""
https://code.fb.com/ai-research/fastmri/
https://code.fb.com/ai-research/fastmri/

Labels: Oleg Zabluda

"""
"""
Google has paid $1 billion for a huge Mountain View business park, the Bay Area’s largest real estate purchase this year.

It is also the second-largest property purchase in the United States this year, eclipsed only by another Google acquisition, the $2.4 billion the company paid for Chelsea Market in Manhattan.

The newly acquired site in Mountain View, where Google has been the primary tenant, is larger than the property that accommodates the company’s Googleplex headquarters a few blocks to the west and also exceeds the size of the parcel across the street where Google is building an iconic “dome” campus that features canopies and tents.
[...]
in the two years since the search giant began to collect properties in downtown San Jose for a proposed transit village, the company has spent at least $2.83 billion in property acquisitions in Mountain View, Sunnyvale, downtown San Jose and north San Jose alone. [...] reaches $3 billion when including the company’s pending purchase in downtown San Jose of several government-owned parcels, along with the minimum value of a big set of surface parking lots that Google intends to buy from Trammell Crow, also downtown near its proposed transit village.

Buying Mountain View’s Shoreline Technology Park gives Google a 51.8-acre site with 12 buildings. The big parcel, which is dominated by one and two-story buildings with ample surface parking, has addresses ranging from 2011 to 2091 Stierlin Court. Aside from Google, a survey of the site indicates that the only other tenant in the complex besides Google is Alexza Pharmaceuticals.

Since its December 2016 purchase nearly two years ago of an old Pacific Bell building near the Diridon Station in San Jose, Google, directly or through affiliates, has spent at least $1.32 billion buying sites in Sunnyvale, $1 billion acquiring properties in Mountain View, $271.8 million in north San Jose and $234 million obtaining parcels in downtown San Jose.

Google’s agreement to pay $109.9 million for government-owned properties in downtown San Jose, as well as the minimum value of $58.5 million for a nearby Trammell Crow-owned site, brings the company’s buying surge in those four markets to $3 billion.
"""
https://www.mercurynews.com/2018/11/26/billion-dollar-deal-google-pays-1-billion-for-huge-mountain-view-business-park/
https://www.mercurynews.com/2018/11/26/billion-dollar-deal-google-pays-1-billion-for-huge-mountain-view-business-park/

Labels: Oleg Zabluda

"""
"""
The Chesapeake–Leopard affair was a naval engagement that occurred off the coast of Norfolk, Virginia, on June 22, 1807, between the British warship HMS Leopard and the American frigate USS Chesapeake. The crew of Leopard pursued, attacked, and boarded the American frigate, looking for deserters from the Royal Navy. Chesapeake was caught unprepared and after a short battle involving broadsides received from Leopard, the commander of Chesapeake, James Barron, surrendered his vessel to the British. Chesapeake had fired only one shot.

Four crew members were removed from the American vessel and were tried for desertion, one of whom was subsequently hanged. Chesapeake was allowed to return home, where James Barron was court martialed and relieved of command.
[...]
The incident outraged the American sense of honor. Americans of every political stripe saw the need to uphold national honor, and to reject the treatment of the United States by Britain as a third class nonentity. Americans talked incessantly about the need for force in response. [...] James Monroe, then a foreign minister acting under instructions from U.S. Secretary of State James Madison, demanded British disavowal of the deed, the restoration of the four seamen, the recall of Admiral Berkeley, the exclusion of British warships from U.S. territorial waters, and the abolition of impressments from vessels under the United States flag. [OZ: британцы послали амеров нах]

many Americans demanded war because of the attack, but President Jefferson turned to diplomacy and economic pressure in the form of the ill-fated Embargo Act of 1807. [...]
while possibly not a direct cause, was one of the events leading up to the War of 1812. [OZ: где амеры огребли люлей, which was a good thing or we'd now have all those stupid Canadians]
"""
https://en.wikipedia.org/wiki/Chesapeake–Leopard_affair
https://en.wikipedia.org/wiki/Chesapeake–Leopard_affair

Labels: Oleg Zabluda

"""
"""
Airbnb’s capitulation underscores the need for Congress to pass the Israel Anti-Boycott Act, which would bar U.S. firms from complying with U.N. boycotts of Israel, like they’re already prohibited from adhering to the Arab League’s boycott. Many U.S. states also have laws prohibiting their pension funds from investing in companies that boycott Israel or territories it administers. State pension boards will likely be looking at Airbnb’s policy before its planned initial public offering next year.
[...] Many states, such as Florida and Alabama, let public employees traveling on official business use Airbnb. These governments should immediately suspend permission to use Airbnb
"""
https://www.wsj.com/articles/airbnbs-anti-israel-hypocrisy-1543175767?emailToken=fd20946a9732562a4fbc72f89b05ac21lJVc62steRNhE5E1646X6PIXEh25+z+hNKA93uIwpxFLS5uDQR5Oz876WeFG%2F1vhoNLqtmlmYEiZYd+qULMeiQ%3D%3D

https://www.wsj.com/articles/airbnbs-anti-israel-hypocrisy-1543175767?emailToken=fd20946a9732562a4fbc72f89b05ac21lJVc62steRNhE5E1646X6PIXEh25+z+hNKA93uIwpxFLS5uDQR5Oz876WeFG/1vhoNLqtmlmYEiZYd+qULMeiQ%3D%3D

Labels: Oleg Zabluda

About Me