Vladimir Polony

It was a great serendipity to meet Eivind Antonsen Segtnan and to support Segtnan.ai effort to practically implement their recent research in hospitals to make disease testing up to 80% more efficient. Btw, their paper just got published in Nature Scientific Reports: https://lnkd.in/dgR56Jj2. Congrats! TL;DR This solution has the potential to save millions when testing for diseases such as Covid-19, HIV, and chlamydia, by grouping and testing samples from multiple individuals together. Segtnan.ai's mobile app integrates AI to tailor risk assessments and employ data-driven methods to optimize group size. Their groundbreaking solution can significantly increase testing capacity and save both time and money in the healthcare system.

11

2 Comments

Love this paper. There are countless low-hanging fruits in deploying LLMs in important fields, including healthcare. I particularly liked this use case. Social Determinants of Health (SDoH) have been close to my heart since my work with RWJF (with Oktawia Wojcik) and Sutter Health (with Sylvia Sudat) to extract them from EHRs many years ago. This is a clear-cut instance where AI is not only effective but better than humans in annotating and surfacing important context for the healthcare provider. #AI #healthcare https://lnkd.in/eGMkjGFQ

22

2 Comments

Today I'm launching the latest weekend hackathon: https://trialsight.ai. It's an #AI agent that watches for the latest updates from the #FDA on #ClinicalTrials.gov. When it finds a trial that matches your personalized criteria, it answers your questions about the trial and delivers each update as a personalized newsletter to your inbox. It's completely free and requires no account registration. Unsubscribe at any time. It's the perfect way to stay on top of the latest clinical trials registered with the FDA whether you're a patient, #lifescience industry analyst, #healthcare worker, or #pharma executive. Check it out, try the preview, and join the waitlist!

19

5 Comments

Our paper, "QGen: On the Ability to Generalize in Quantization Aware Training," (https://lnkd.in/gXXvS_9Q) is now available on Arxiv. This research delved into the generalization capabilities of quantized models, a relatively unexplored domain, examining them through the lens of sharpness in the loss landscape during optimization. I found the experiments and mathematical proofs particularly fulfilling. We're hopeful for acceptance at the conference to which it's been submitted. 🤞

47

This week, I was discussing Bayes’ theorem with one of the scientists on my team. It is one of my favorite statistical theory, and is often counterintuitive. Here is an example: Imagine a rare disease X, affecting 1 in 1000 people. Assume, there's a diagnostic test with 99% accuracy in detecting the disease when present, with a 2% false positive rate. Let’s say someone (Person A) takes the test, and it comes back positive. What's the likelihood that they actually have the disease? At first glance, one might think that the probability of having the disease, given a positive test result, would be high – especially since the test is 99% accurate with only a 2% false positive rate. But is that the case? Bayes’ theorem can give the exact answer, here is the number crunching: Prob (Having the disease X) = P(A) = 0.1% Prob (Not having the disease X) = P(A’) = 1- P(A) = 99.9% Prob (Getting a positive test result, when the disease is present) = P(B∣A) = 99% Prob (Getting a positive test result, when the disease is not present) = P(B∣A’) =2% From law of total probability, Prob (Getting a positive test result) = P(B)=P(B∣A) ×P(A)+P(B∣A′) ×P(A′) = 2.1% From Bayes’ theorem, Prob (Having the disease, when the test is positive) = P(A∣B) = P(B∣A) ×P(A)​ / P(B) = 0.04726 = 4.71% So, despite the positive test result, the likelihood that person A actually has the disease X is only 4.7% (in other words, the likelihood that person A does not have the disease X even with a positive test result is quite high, at 95.3%). This is what makes Bayes' Theorem so fascinating! #DataScience #Statistics #Bayes’Theorem

25

1 Comment

The Open Medical LLM Leaderboard on Hugging Face dropped last Friday (thanks Geoff Kwitko for the heads up), and it aims to track, rank and evaluate the performance of large language models (LLMs) on medical question answering tasks (see link below). It evaluates LLMs across a diverse array of medical datasets, including MedQA (USMLE), PubMedQA, MedMCQA, and subsets of MMLU related to medicine and biology. The leaderboard offers a comprehensive assessment of each model's medical knowledge and question answering capabilities. The datasets cover various aspects of medicine such as general medical knowledge, clinical knowledge, anatomy, genetics, and more. They contain multiple-choice and open-ended questions that require medical reasoning and understanding. More details on the datasets can be found in the "LLM Benchmarks Details" section in the link. Ps. Good to know that the GPT-4 model is leading the way (it is what I use on the Medical Coding and Documentation agent). Digital Human Assistants #llmhealthcare

8

This is a wonderful paper. It shows how clever design (with simplicity and elegance) in architecture can improve the performance of transformers in arithmetic, rather than by simply throwing in more data and compute. https://lnkd.in/gZNnifc3

14

I'll do even a braver poll than last, but only because of this Eric Topol, MD article in Science Magazine :) Multimodal large AI models will be used in medical forecasting (disease risk, onset, trajectory) A) now B) soon but at population level only C) sooner than in diagnostics D) later than in diagnostics E) never https://lnkd.in/gGBHSptv

7

John Snow Labs new JSL-MedLlama-3-8B-v3.0 #Medical #LLM is the first LLM under 70B parameters to pass 80% on the Open Medical LLM Leaderboard. With an average score of 80.69 across 9 benchmarks, it's currently 3 points above the next-best model - and also outperforms GPT-4 and Med-PaLM2 on the MMLU (college-level question answering) benchmarks. Leaderboard: https://lnkd.in/ghvWHwZ3 #Healthcare #NLP & #LLM Library (for commercial use): https://lnkd.in/gc2DYMJQ

91

3 Comments

Very insightful work by Anthropic’s interpretability team. And an amazing paper, with outstanding writing and figures. The idea is very simple: interpret LLMs by leveraging sparse autoencoders as surrogate models of the MLP of transformer blocks, which allow one to disambiguate the superposition of features captured by a single neuron. A simple idea, but a very careful and complex execution, as it is often the case in our line of work. The paper goes into many details and provide a large array of insights, although the gist of the implementation remains obfuscated due to the closed source nature of Claude. Too bad, because this is the kind of work that we need to better understand and eventually trust LLMs. This is demonstrated by the authors in the section ‘Influence on Behavior’, where they show that clamping some features to either high or low value during inference is “remarkably effective at modifying model outputs in specific, interpretable ways”. Hopefully this kind of work is going to be replicated and generalized to open-weights models, such that we have new ways to steer their behavior. https://lnkd.in/eVym7f_f #interpretability #xai #explainableai #steerableai #anthropic #claude #anthropic

3

NEW AI MATH BOOK Excited to announce that I just published a book: "The Foundational Mathematics of Artificial Intelligence." Hardcover available now on Amazon. Other formats are in the works from publisher, Symmetry Seed Books. Essential AI algorithms and their math explained in 20 chapters and 383 pages. International book tour details are pending. Please reach out if you would like me to come speak about the book and about AI at your institution. BOOK CHAPTERS: 1. The True History of AI 2. The Building Blocks (Calculus & Linear Algebra) 3. Linear Regression 4. Logistic Regression 5. Constrained Optimization 6. Support Vector Machines 7. Fourier Transforms 8. Eigenvalue Decomposition 9. Singular Value Decomposition 10. Principal Component Analysis 11. Neural Networks 12. Deep Reinforcement Learning 13. Generative Adversarial Networks 14. Variational Autoencoders 15. Diffusion Denoising Probabilistic Models 16. Word Embedding: Word2Vec 17. Attention Mechanism 18. Transformer Architecture 19. Contrastive Learning Image Pretraining (CLIP) 20. Large Language Models Artificial Intelligence Machine Learning Bayo (Olubayo) Adekanmbi PhD Jon Reifschneider Duke AI Master of Engineering AIMed Peter Blair Jürgen Schmidhuber U(M) Mathematics University of Alabama at Birmingham Duke University School of Medicine Howard University Duke University DSNai - Data Science Nigeria Black in AI Karim Beguir Ndubuisi Ekekwe Michael F. Chiang Theodore Leng, MD, MS Judy Gichoya stephen moore Dr Nick Bradshaw Tobi Olatunji MD Deep Learning Indaba Daniela Ferrara, MD PhD FASRS Daniel Ting Hisani (Horne) Madison, PhD, MPH Jana Delfino Pearse Keane https://lnkd.in/gM6wKqqU

323

87 Comments

New Mindset Needed for Large Language Models - https://buff.ly/3XbN4vp #AI #ML #LLMs #genAI

A new research paper by our incredible teams at Google Research and Google DeepMind shows the strong performance of Med-Gemini across a variety of multimodal tasks, including  2D and 3D imaging in radiology, histopathology, ophthalmology, and dermatology, and genomics. Med-Gemini ranked best in class on chest X-ray and CT scan report generation, and is able to perform genomics-based disease prediction and mortality risk more accurately than a standard linear polygenic risk score (PRS) based approach. This work couldn't have been possible without close collaboration with our partners, including a panel of specialists who assessed and rated our models as part of this research. These promising results showcase the potential of LLMs to transform medicine as we know it. https://lnkd.in/dHx8_DYh

279

5 Comments

Small models becoming smarter! Phi-3 reinforces the importance of data quality.

24

First test email from Zeeq.ai came through overnight. Zeeq (rhymes with "seek") is a personalized AI agent that watches ClinicalTrials.gov for updates of #clinicaltrials for trials of interest and generates answers to your questions delivered straight to your inbox whenever a trial gets added or updated. Had a few hiccups with reading the change feed from the FDA, but looks like it's almost there and getting closer to full rollout (a few more days of observing, I think!). Whether you are a #patient, #pharma exec, #biotech industry analyst, or academic researcher, Zeeq can help you stay on top of the most relevant trials of interest with personalized questions and answers. Free to sign up at https://zeeq.ai - no usernames or passwords, no accounts.

11

3 Comments

2023 was a breakout year for investment in generative AI startups, with equity funding of almost $22B across 426 deals. The 5 largest rounds all went to companies focused on core generative AI infrastructure: * OpenAI, AI poster child and maker of ChatGPT ($10B corporate minority) * Inflection AI, which focuses on human-computer interfaces ($1.3B Series B) * Anthropic, an AI model developer and research outfit ($1.8B across 2 corporate minorities) * Databricks, a data integration and analytics platform ($504M Series I) * Aleph Alpha, Germany-based LLM developer ($500M Series B) When coupled with the pace of change and the compression of time and cost, makes anyone wonder, given the apparent speed of the commoditization of these LLM engines, what will the ROIC on funds deployed actually look like over time?

1

🔬🖥️ Google Cloud is teaming up with Bayer to revolutionize radiology through AI-enhanced tools, promising to significantly alleviate the workload for radiologists and improve patient care. This collaboration leverages Google's cloud computing prowess and Bayer's expertise in healthcare to develop a platform focused on radiology applications, such as CT scans, ultrasounds, and MRIs. Set to launch later this year in the U.S. and Europe, the platform will support developers in securely experimenting with and analyzing massive volumes of healthcare data, constituting about 90% of all healthcare data. It will also include features for AI-powered searches of scientific papers and regulatory documents, assisting in swiftly acquiring necessary regulatory approvals. https://lnkd.in/dUMeuguN #HealthcareInnovation #AIinHealthcare #Radiology

55

5 Comments

I spoke today with computational bio expert Jefferson Parker, PhD (at NullSet Informatics Solutions) for our State of AI in Biotech report. He shared some interesting career advice for those working in AI: 9 out of 10 leaders in biotech want AI in their scientific development, but do not know WHY. 1 out of 10 DO know why. They probably have the best data, and that is a blessed place to be and develop in your career. Always ask why leaders want AI and much more will be revealed. And for leaders, understanding why you want AI (beyond the hype and promises) will lead to a team culture that prioritizes data and let's the best talent flourish. It's win-win.

24

3 Comments