“I was Steve's advisor at the Harvard Kennedy School, though he did not need much advising. He has one of the quickest and most imaginative minds ever to grace this place. He had then, and certainly has now, a remarkable ability to blend (and create) cutting-edge technologies with a creative and contrarian approach to the social sciences. I predicted then, and believe now, that Steve Purpura will be among the most important scholars of his generation.”
About
Global leader with a focus on improving data, systems, and algorithms for machine…
Activity
-
The latest perk I’m hearing for AI startups is that they own their own Autel fast DC charger for employee use when they’re at the office.
The latest perk I’m hearing for AI startups is that they own their own Autel fast DC charger for employee use when they’re at the office.
Posted by Stephen P.
-
Some people working on AI are whispering about a major gain in a LLM’s ability to translate English to mathematical formulas that actually work…
Some people working on AI are whispering about a major gain in a LLM’s ability to translate English to mathematical formulas that actually work…
Posted by Stephen P.
-
In a Twitter discussion about a month ago with Big Guns Pete and Joe Heitzeberg, I mentioned that I'm not seeing corporations (non-magnificent 7, non…
In a Twitter discussion about a month ago with Big Guns Pete and Joe Heitzeberg, I mentioned that I'm not seeing corporations (non-magnificent 7, non…
Shared by Stephen P.
Experience & Education
-
Autonomous Data Management LLC
View Stephen’s full experience
See their title, tenure and more.
or
Volunteer Experience
-
Advisory Counsel Member, Open Data Initiative
New York City Fire Department
- 1 year 7 months
Science and Technology
Great people in the Fire Department allowed me the opportunity to give back to the community by helping them craft ways for citizens to engage with community data to save lives.
Publications
-
Statistically adaptive learning for a general class of cost functions (SA L-BFGS)
arXiV
We present a system that enables rapid model experimentation for tera-scale machine learning with trillions of non-zero features, billions of training examples, and millions of parameters. Our contribution to the literature is a new method (SA L-BFGS) for changing batch L-BFGS to perform in near real-time by using statistical tools to balance the contributions of previous weights, old training examples, and new training examples to achieve fast convergence with few iterations. The result is, to…
We present a system that enables rapid model experimentation for tera-scale machine learning with trillions of non-zero features, billions of training examples, and millions of parameters. Our contribution to the literature is a new method (SA L-BFGS) for changing batch L-BFGS to perform in near real-time by using statistical tools to balance the contributions of previous weights, old training examples, and new training examples to achieve fast convergence with few iterations. The result is, to our knowledge, the most scalable and flexible linear learning system reported in the literature, beating standard practice with the current best system (Vowpal Wabbit and AllReduce). Using the KDD Cup 2012 data set from Tencent, Inc. we provide experimental results to verify the performance of this method.
Other authors -
Patents
-
Training a machine learning model
Issued US 9454733
See patentOptimized methods of training machine learning models
-
Selecting a training strategy for training a machine learning model
Issued US 9449283
See patentMethods for optimizing the selection of a training strategy for machine learning.
-
Re-training a machine learning model
Issued US US 13/969,193
See patentMethods, systems, and apparatus, including computer programs encoded on computer storage media, for training machine learning models. The models can include models for predicting a next transaction price or a next transaction price direction for one or more financial products, for classifying particular debit or credit card transactions as likely being anomalous or fraudulent or not, or for classifying particular financial claims processing transactions, e.g., insurance, health care, or…
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training machine learning models. The models can include models for predicting a next transaction price or a next transaction price direction for one or more financial products, for classifying particular debit or credit card transactions as likely being anomalous or fraudulent or not, or for classifying particular financial claims processing transactions, e.g., insurance, health care, or employee expense claims transactions, as likely being anomalous or fraudulent or not.
-
Managing state information across communication sessions between a client and a server using a stateless protocol
Issued US US 2004/0181598 A1
See patentThis patent reinvented web cookies to allow browser applications to maintain authentication information on web browsers that had disabled cookies. It allowed for behavioral tracking of web site users even when their browser cookies were disabled. I originally invented this technology as part of a dialogue with 13 state attorneys general over the effectiveness of laws restricting cookie implementation in the browser.
-
Method and system for protecting internet users' privacy by evaluating web site platform for privacy preference policy
Issued US 7614002
See patentW3C P3P implementation in Internet Explorer
-
Security component for a computing device
Issued US 7058978
See patentComputers have a need to have an internal firewall that allows application layer inspection and intermediation. This is the patent for that.
-
Method and system for protecting internet users' privacy by evaluating web site platform for privacy preferences policy
Issued US 6959420
See patentW3C P3P implementation in Internet Explorer
-
Method and system for authentication and single sign on using cryptographically secured cookies in a distributed computing environment
Issued US 6421768
See patentBanks, banking service providers and other third parties that need to seamlessly transition customers between web sites needed a mechanism to allow this to occur.
Honors & Awards
-
Amazon Web Services Commercial Employee of the Quarter, Q1 2019
Amazon Web Services
-
Finalist, EY Entrepreneur Of The Year® 2015 in the Pacific Northwest
Ernst & Young LLP
Finalist, EY Entrepreneur Of The Year® 2015 in the Pacific Northwest
-
Nominated, Best Paper Award
Computer Human Interaction (CHI) Conference
-
Cornell Department of Computer Science Outstanding Teaching Award
-
-
Yahoo! Graduate Student Teaching Award
Yahoo!
Awarded to the best graduate student instructor in Computer Science at Cornell University, Ithaca campus.
Recommendations received
1 person has recommended Stephen
Join now to viewMore activity by Stephen
-
6.24.24 - Dave's Weekly Software Engineering Job Ad Count This week’s job ad numbers show a demand spike across all categories (SDEs, Product Mgrs,…
6.24.24 - Dave's Weekly Software Engineering Job Ad Count This week’s job ad numbers show a demand spike across all categories (SDEs, Product Mgrs,…
Liked by Stephen P.
-
Apple released 20 new CoreML models on Huggingface this morning, some of which significantly reduce power/compute requirements for edge ML.
Apple released 20 new CoreML models on Huggingface this morning, some of which significantly reduce power/compute requirements for edge ML.
Shared by Stephen P.
-
I was waiting for this paper to come out. There are proposed formal academic definitions for bullshit that are actively debated and ChatGPT's output…
I was waiting for this paper to come out. There are proposed formal academic definitions for bullshit that are actively debated and ChatGPT's output…
Shared by Stephen P.
-
That product managers ship a feature that would subject you to total lockout of your account by SIM Swap is borderline dereliction of duty to users
That product managers ship a feature that would subject you to total lockout of your account by SIM Swap is borderline dereliction of duty to users
Liked by Stephen P.
-
Continuing the series of posts about research papers on LLMs that do not require massive Matrix Multiplies. This paper shows a power efficiency for…
Continuing the series of posts about research papers on LLMs that do not require massive Matrix Multiplies. This paper shows a power efficiency for…
Shared by Stephen P.
-
Google loves open source. Great partnership with Jerry Liu and LlamaIndex team and looking forward to deepen our work together. Thanks Erwan Menard ,…
Google loves open source. Great partnership with Jerry Liu and LlamaIndex team and looking forward to deepen our work together. Thanks Erwan Menard ,…
Liked by Stephen P.
-
The reduction of matrix multiplies in LLMs is one of the most interesting areas of research right now because it enables much cheaper deployment of…
The reduction of matrix multiplies in LLMs is one of the most interesting areas of research right now because it enables much cheaper deployment of…
Shared by Stephen P.
-
We didn't want to make a big deal of Bloomberg Beta turning 11 today, but someone on the team... had other plans. Grateful for every year that…
We didn't want to make a big deal of Bloomberg Beta turning 11 today, but someone on the team... had other plans. Grateful for every year that…
Liked by Stephen P.
-
The new Intel Xeon 6700 performance per watt is getting close to ARM levels. The eCores are good at lots of database and media VMs on a box. We're…
The new Intel Xeon 6700 performance per watt is getting close to ARM levels. The eCores are good at lots of database and media VMs on a box. We're…
Posted by Stephen P.
-
From a few polls I've seen recently, 60% of product managers are embedding their PRDs in ChatGPT.
From a few polls I've seen recently, 60% of product managers are embedding their PRDs in ChatGPT.
Posted by Stephen P.
-
A new paper and source code depot for a new quantization method that improves LLM serving by increasing GPU efficiency at quantization. Performance…
A new paper and source code depot for a new quantization method that improves LLM serving by increasing GPU efficiency at quantization. Performance…
Shared by Stephen P.
-
Why is an Army guy in Warships International Fleet Review? To discuss the importance of the conflict in the Ukraine and the case for the West to keep…
Why is an Army guy in Warships International Fleet Review? To discuss the importance of the conflict in the Ukraine and the case for the West to keep…
Liked by Stephen P.
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Stephen P. in United States
-
Stephen P
Recruitment Consultant
-
Stephen P
Software Engineer - Huntsville
-
Stephen P
Revenue Management
-
Stephen P.
Chairman at Qatari Diar Americas Inc.
-
Stephen P.
549 others named Stephen P. in United States are on LinkedIn
See others named Stephen P.