Michael Barnathan
New York City Metropolitan Area
2K followers
500+ connections
About
Seasoned generalist software engineering and ML leader. I've been building software since…
Articles by Michael
-
What Business to Start? The Four Stages of Technical Ecosystems
What Business to Start? The Four Stages of Technical Ecosystems
By Michael Barnathan
-
Managing a Distributed Team in the Age of Coronavirus
Managing a Distributed Team in the Age of Coronavirus
By Michael Barnathan
Activity
-
Been quietly contemplating all the new technologies that have emerged over the last 17 years I've worked in AI. What's working. What's missing…
Been quietly contemplating all the new technologies that have emerged over the last 17 years I've worked in AI. What's working. What's missing…
Liked by Michael Barnathan
-
We are hiring multiple Machine Learning engineers for Anthropic's Trust & Safety team in SF and NYC! - https://lnkd.in/gX8Vk27W Please feel free to…
We are hiring multiple Machine Learning engineers for Anthropic's Trust & Safety team in SF and NYC! - https://lnkd.in/gX8Vk27W Please feel free to…
Liked by Michael Barnathan
Experience
Education
-
Temple University
-
Activities and Societies: Researcher (and the system administrator) at the Data Engineering Laboratory, member of the ACM, Member of the Golden Key Honor Society. My research had significant biomedical applications (I coauthored with two investigators who later went on to CRISPR HIV out of animals) and I would like to continue exploring this field.
I defended my thesis on using tensors to perform machine learning on images, "Mining Complex High-Order Datasets", on April 23, 2010 (frontrunning Tensorflow by 5 years). At Temple, I was completely funded by a 4-year Temple University Fellowship and a Dean's Scholarship.
I applied high-order versions of SVD to MRIs to extract latent spaces (~embeddings) after preprocessing with novel wavelet-based techniques to capture neighborhoods. I've also applied these methods to graph mining and…I defended my thesis on using tensors to perform machine learning on images, "Mining Complex High-Order Datasets", on April 23, 2010 (frontrunning Tensorflow by 5 years). At Temple, I was completely funded by a 4-year Temple University Fellowship and a Dean's Scholarship.
I applied high-order versions of SVD to MRIs to extract latent spaces (~embeddings) after preprocessing with novel wavelet-based techniques to capture neighborhoods. I've also applied these methods to graph mining and recommendation engines, and have made significant extensions to (+ wrote the first opensource implementation of) the WaveCluster clustering algorithm.
Also used VQ keyblock encodings to create sequences of image patches known as "visual vocabularies", allowing computer vision problems to be addressed with text and sequence mining techniques such as financial timeseries modeling and genetic sequence analysis. These techniques later found their way into vision transformers and latent diffusion models. -
-
Activities and Societies: See Ph. D. section above.
I earned my Master's in the course of my career as a Ph. D. student at Temple. By taking an accelerated courseload (4 graduate courses each semester and a research project in the summer), I was able to complete this degree in one year.
I learned how to read MRIs and detailed anatomy of the upper half of the body (down to about the aortic arch) from this experience. I would later accurately diagnose an extremely rare, thankfully benign, subcentimeter tumor initially missed by two…I earned my Master's in the course of my career as a Ph. D. student at Temple. By taking an accelerated courseload (4 graduate courses each semester and a research project in the summer), I was able to complete this degree in one year.
I learned how to read MRIs and detailed anatomy of the upper half of the body (down to about the aortic arch) from this experience. I would later accurately diagnose an extremely rare, thankfully benign, subcentimeter tumor initially missed by two experienced subspecialty-trained radiologists on my own scans. (Honestly, it had a textbook signal and I have no idea how they both missed it. The diagnosis was confirmed.)
Master's Project: "A Web-Accessible Framework for Automated Storage and Texture Analysis of Biomedical Images".
See the Ph. D. section for additional information. -
-
Activities and Societies: Member of the Science, Technology, and Engineering advisory council (Sep. 2004 - May 2007), Phi Eta Sigma, Lambda Sigma Tau, Kappa Mu Epsilon, and Omicron Delta Kappa honor societies, National Dean’s List. President (2005-2006) and Vice President (2004-2005) of the Monmouth University ACM chapter, Officer (2005-2006) of the Lambda Sigma Tau honor society.
Highest GPA in the Class of 2006 (recipient of a $5000 award for that). Also a co-recipient of the 2006 Excellence in Computer Science Award.
Graduated Summa Cum Laude with a GPA of 3.96 and membership in the Dean's List every semester.
Recipient of a $6000 Monmouth University Scholarship and the Dr. Harold Jacobs Scholarship for Excellence in Science, Technology, or Engineering (twice). Published my first AI article as an undergraduate, on agent based systems.
Publications
-
A Wearable Accelerometer System for Unobtrusive Monitoring of Parkinson Diease Motor Symptoms
BIBE 2014
This presentation was given based on the work that we did classifying Parkinson's Disease symptoms at BioMotion Suite.
-
Mammographic Segmentation Using WaveCluster
Algorithms
Sole author. Barnathan, M. Mammographic Segmentation Using WaveCluster. Algorithms. 2012; 5(3):318-329.
I've become somewhat of an expert on the WaveCluster algorithm, since I authored the only publicly accessible implementation of it (as of mid-2012) and made a number of extensions to the basic algorithm which greatly increase its power. -
TWave: High-Order Analysis of Functional MRI
NeuroImage
Published in Neuroimage. 2011 Sep 15;58(2):537-48
-
Mining Complex High-Order Datasets
Dissertation, Temple University
My doctoral dissertation introduces TWave, the first comprehensive framework for high-order classification, clustering, compression, feature extraction, and latent concept discovery. Pioneering the use of tensor factorizations in biomedical image analysis, the principal application was in mining motor task fMRI data. Key insights include the formulation of grid clustering (on a grid of arbitrary dimensionality and order) as a locality-sensitive instance of latent semantic analysis, the fusion…
My doctoral dissertation introduces TWave, the first comprehensive framework for high-order classification, clustering, compression, feature extraction, and latent concept discovery. Pioneering the use of tensor factorizations in biomedical image analysis, the principal application was in mining motor task fMRI data. Key insights include the formulation of grid clustering (on a grid of arbitrary dimensionality and order) as a locality-sensitive instance of latent semantic analysis, the fusion of wavelets with tensor factorizations to achieve a methodology which is sensitive to both global patterns and local neighborhoods at multiple resolutions, and the extension of WaveCluster into a versatile high-order, global, real-valued image clustering algorithm called TWaveCluster dominated by WaveCluster's linear runtime. Additional results include a novel tensor-theoretic formulation of the multidimensional separable wavelet transform which permits application to datasets of arbitrary order, a margin-sensitive variant of WaveCluster using Lloyd iteration to deform the grid cells prior to analysis (eliminating the potentially massive partial volume effect inherent in the naive WaveCluster algorithm), and the development of novel image analyses, including a texture-based computer-assisted diagnosis system achieving up to 89% accuracy in classification of brain tumor MRIs. From a 9.2 GB 6th-order digital-opposition task fMRI dataset, TWave derived subject handedness as a latent concept without human intervention, was capable of automatically identifying and distinguishing spatiotemporal activation patterns in the frontal and motor regions of the brain corresponding to planning to move and actual movement, unambiguously demonstrated the existence of ipsilateral motor activation in our dataset, and yet consumed two orders of magnitude less time and space vs. existing methods of image analysis.
-
A Representation and Classification Scheme for Tree-Like Structures in Medical Images: Analyzing the Branching Pattern of Ductal Trees in X-ray Galactograms
IEEE Transactions on Medical Imaging
I am the second author on this paper. Among my first publications in the biomedical domain, this paper presents a methodology for topological analysis of the breast ductal network using Prufer and depth-first tree encodings with applications to diagnosis of breast cancer and other radiological abnormalities of the breast (with high discriminative value, as these conditions often distort the ductal architecture and thus radically alter the branching patterns, sometimes even before a gross lesion…
I am the second author on this paper. Among my first publications in the biomedical domain, this paper presents a methodology for topological analysis of the breast ductal network using Prufer and depth-first tree encodings with applications to diagnosis of breast cancer and other radiological abnormalities of the breast (with high discriminative value, as these conditions often distort the ductal architecture and thus radically alter the branching patterns, sometimes even before a gross lesion is visible on mammography). The trees are manually segmented in this work; another paper I co-authored three years later would introduce an automatic segmentation method using a set of hybrid local features.
-
Analyzing Tree-Like Structures in Biomedical Images Based on Texture and Branching: An Application to Breast Imaging
Proceedings of the International Workshop on Digital Mammography 2008
In this work I fuse texture and topological analysis of branching patterns of the breast ductal network to match datasets of galactograms and mammograms. The methods used to assess branching were developed in my previous paper, but the novelty in this paper is nontrivial.
-
A Texture-Based Methodology for Identifying Tissue Type in Magnetic Resonance Images
Proceedings of the 8th International Symposium on Biomedical Imaging
I'm particularly fond of the results of this paper, which utilizes keyblock texture learning (or "image patches", as they're now called) and vector quantization to distinguish between 11 types of normal and abnormal brain tissue in T1 and T2 MRIs of a transgenic mouse with a large medulloblastoma, including vascularization, peritumoral edema, and tissue necrosis - these imaging features are important as they can be used to grade human gliomas.
(I generally have some objections to using…I'm particularly fond of the results of this paper, which utilizes keyblock texture learning (or "image patches", as they're now called) and vector quantization to distinguish between 11 types of normal and abnormal brain tissue in T1 and T2 MRIs of a transgenic mouse with a large medulloblastoma, including vascularization, peritumoral edema, and tissue necrosis - these imaging features are important as they can be used to grade human gliomas.
(I generally have some objections to using transgenic animal tumor models in imaging studies when we're unfortunate enough to possess large volumes of human tumor imaging data, but this aspect of the work was out of my hands)
I had no idea how close I was working to the ideas that would lead to vision transformers here.
Patents
-
System for automatic semantic enrichment and augmentation of unstructured web content
Filed 63436627
Automatic construction of an ontology of concepts on a web page. Machines are going to build the Semantic Web from the unstructured web, not humans painstakingly annotating things in OWL.
Projects
-
Stealth Social Agent Project
Pushing the envelope on Agents, to the point that I've become an upstream contributor to agentops and CrewAI and have created an abstraction layer that unifies CrewAI and Autogen.
-
Project Glow Layer: The Semantic Web
-
I've found a way to deliver the semantic web to everyone without needing to annotate anything. Patent pending. Spiked this out but elected not to continue with it when browser manufacturers started moving in the same direction.
-
The Crucible
-
A highly scalable multi-objective learning system built on completely different fundamentals from conventional machine learning systems. Bit of a moonshot. Needs some post-ChatGPT modifications.
-
Babytracker Multiples for Alexa
-
Alexa integration for Babytracker, extended with support for multiple babies. Because while I'm doing all of this awesome stuff, I'm also a proud twin daddy.
-
Sharpshooter
-
A multi-step cryptocurrency arbitrage bot written using Python asyncio. Capable of identifying not only individually profitable arbitrage trades, but profitable *sequences* of trades, even those with steps that individually lost money. Since I could identify setups that single-step arbitrage bots (i.e. almost all of them) missed, I made some nice money on this in 2017 and stayed in the game much longer than most. Eventually the crypto markets became efficient enough that even this stopped…
A multi-step cryptocurrency arbitrage bot written using Python asyncio. Capable of identifying not only individually profitable arbitrage trades, but profitable *sequences* of trades, even those with steps that individually lost money. Since I could identify setups that single-step arbitrage bots (i.e. almost all of them) missed, I made some nice money on this in 2017 and stayed in the game much longer than most. Eventually the crypto markets became efficient enough that even this stopped working.
-
SmartWink
-
A bridge between the SmartThings, Wink, and Lutron smart home ecosystems, featured on HackADay.
-
Parkinson's Knee Brace
-
Created a knee brace that could detect Parkinsonian tremor and dyskinetic symptoms, designing an embedded machine learning library and training it on an Arduino with a three-axis accelerometer. The system achieved roughly 80% sensitivity in tests with nearly no false positives. This work generated a publication.
-
Stochastic Vocabularies
-
I generalized a computer vision (texture analysis) technique known as keyblock encoding to domains outside of the visual one, modeling such diverse phenomena as audio waveforms, financial timeseries, and genomes by creating "vocabularies", quantizing signals to these vocabularies, and analyzing the transitions between "words" that represent the quantized signals using stochastic methods.
If this sounds familiar, it's because ViT and other transformers now use the same technique to reduce…I generalized a computer vision (texture analysis) technique known as keyblock encoding to domains outside of the visual one, modeling such diverse phenomena as audio waveforms, financial timeseries, and genomes by creating "vocabularies", quantizing signals to these vocabularies, and analyzing the transitions between "words" that represent the quantized signals using stochastic methods.
If this sounds familiar, it's because ViT and other transformers now use the same technique to reduce images and other media to sequences. -
Metasquared
-
After AOL took down a game by MetaTools called Metasquares, I singlehandedly developed a popular replacement with an AI opponent and full multiplayer online play. This used the REST paradigm several years before it was formally invented, as well as a fairly sophisticated implementation of alpha-beta pruning. I was about 12 at the time.
-
Final Aegis
-
Final Aegis was a highly customized Ultima Online shard which I ran using Sphere Server under the GM name "Rudenid". I became known as one of 3 or so top experts in the Sphere Scripting community due to the "impossible" things I would code into the game experience, such as floating cities, airships, mind control spells, quests, and time mages.
-
Quantile
-
I built a portfolio screener which used a variety of financial risk metrics (e.g. VaR, Sharpe, Treynor, Drawdown) as well as ML signals to evaluate the health of a portfolio. 2007 was an interesting time to do this.
Honors & Awards
-
Temple College of Science and Technology Dean's Scholarship
Temple University College of Science and Technology
A supplemental boost to my stipend over the amount provided by my University Fellowship.
-
Temple University Fellowship - 4 years
Temple University
My tuition and stipend as a graduate student at Temple University were covered by a 4 year University Fellowship.
-
Monmouth University Alumni Association Academic Achievement Award for Highest GPA in the Class of 2006
Monmouth University
I received a $5000 cash award at the commencement ceremony for graduating Monmouth University with the highest GPA in my class.
-
Departmental Computer Science Award
Monmouth University Department of Computer Science
I received one of two departmental awards for excellence in Computer Science given out in my year before graduating.
-
9 extra Google peer/spot bonuses
Other Googlers whom I've significantly helped.
In my two years at Google, I received eight peer bonuses from other Googlers whom I've helped across a multitude of teams, and one spot bonus (a larger award, comparable to a small extra annual bonus) from the head of release engineering in NYC. The spot bonus involved driving a cross-team effort across three teams even though I was an IC.
Languages
-
English
Native or bilingual proficiency
-
Spanish
Limited working proficiency
-
Chinese
Elementary proficiency
Recommendations received
7 people have recommended Michael
Join now to viewMore activity by Michael
-
🚀 Personal News! 🚀 I’m thrilled to share that I’ve joined OpenAI as its first internal Global Data Protection Officer! Why OpenAI? Anyone who…
🚀 Personal News! 🚀 I’m thrilled to share that I’ve joined OpenAI as its first internal Global Data Protection Officer! Why OpenAI? Anyone who…
Liked by Michael Barnathan
-
After leaving OpenStore last Tuesday, this week I shift to working full time on my property management company, Hogar Management, while continuing to…
After leaving OpenStore last Tuesday, this week I shift to working full time on my property management company, Hogar Management, while continuing to…
Liked by Michael Barnathan
-
🏗️ Announcing Trunk Tools $20M Series A! 🏗️ I’m excited to finally share that Trunk Tools has just secured $20 million in Series A funding, led by…
🏗️ Announcing Trunk Tools $20M Series A! 🏗️ I’m excited to finally share that Trunk Tools has just secured $20 million in Series A funding, led by…
Liked by Michael Barnathan
-
I made it to the director level at Alphabet without working more than 40 hours per week. I did this with a very simple 2-step time management…
I made it to the director level at Alphabet without working more than 40 hours per week. I did this with a very simple 2-step time management…
Liked by Michael Barnathan
-
This is likely my last post on this topic (unless something even more revolutionary happens). If you are going to stay in the bunker, don't tell me I…
This is likely my last post on this topic (unless something even more revolutionary happens). If you are going to stay in the bunker, don't tell me I…
Liked by Michael Barnathan
Other similar profiles
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 More