Henry Luckhoo

Kolmogorov–Arnold Networks seem like a paradigm shift in ML architectures. Curious to see how they evolve. "While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights"). KANs have no linear weights at all -- every weight parameter is replaced by a univariate function parametrized as a spline. " https://lnkd.in/gdPk_qCQ

29

The next part of my Causal AI series is up on Towards Data Science, this time exploring how Linear Programming can be used to optimise treatment strategies: https://lnkd.in/eb2ErFgi

46

👉🏼 Reliability of large language models for advanced head and neck malignancies management: a comparison between ChatGPT 4 and Gemini Advanced 🤓 Andrea Lorenzi 👇🏻 https://lnkd.in/eSW4Admy 🔍 Focus on data insights: - ChatGPT 4 generally outperformed Gemini Advanced in adherence to guidelines and comprehensive treatment planning. - Higher AIPI scores were observed with ChatGPT 4, indicating better overall performance. 💡 Main outcomes and implications: - Both large language models showed potential in aiding multidisciplinary management of head and neck oncology. - Discrepancies in critical areas emphasize the need for further refinement and validation against current clinical standards. 📚 Field significance: - Growing role of AI in enhancing clinical decision-making. - Necessity for continuous updates and validation to fully integrate AI into healthcare practices. 🗄️: [#largeLanguageModels #headAndNeckMalignancies #AIclinicalDecisionMaking]

1

1 Comment

A nice and brief overview of the Yellowbrick Data platform by Intellyx following my chat with them this week. https://lnkd.in/eRZQhy5U

16

At the start of my career I would often prioritise the ML algorithm I wanted to use, over the data itself. With time, experience, and mistakes - I've completely flipped the script. Here is my 8-step data preparation checklist to ensure your ML model is as robust and performant as possible: ✅ Missing values - how should they be processed? ✅ Duplicate & low-variation data - can this be removed? ✅ Incorrect & irrelevant data - how do we identify it? ✅ Categorical data - what encoding technique fits best? ✅ Outliers - could they cause issues? ✅ Feature Scaling - is this necessary? ✅ Feature Engineering & Selection - can we help the model learn? ✅ Testing & Validation - which approach makes sense? What would you add? #datascience #analytics #data #datascienceinfinity

60

4 Comments

This week I have been thinking about security models with generative AI. Have a read and let me know your thoughts in the comments section! #MethodsForGood #ai #generativeai #security #cybersecurity #responsibleai# #dataprotection

9

“Explain options trading in simple terms” while I’m on the tube with no internet 🛜 Here’s a very awkward demo on the London Underground of me using #opensource AI tooling (Ollama and Phi-3) to converse in natural language without an internet connection. Why is this important? As models get smaller, our devices get more powerful and all this compute happens locally (your devices - computers, phones, toasters?) we will begin to see new ways to work with models in new places. Here an example: A education application that could be used on your commute that tailors itself to you as you go. Currently that requires internet but with local on device models you could that anywhere, anytime and for free* Phi-3 is “tiny” and could fit on a phone. Its been done before. Apple and Google want to control the model on our phones but we shall see if this actually happens. An OpenAI phone? Maybe but I doubt it. Anyways just thought I’d try this and share it here for impact. We are moving towards on device AI compute. What a time to be alive.

18

5 Comments

💡 personal take on 💢 Ai2 Tülu 3, an open-data & open-pipelines LLM. Hugging Face and the open LLM community will benefit a lot from this. CC: Lewis Tunstall, Jeremy Howard, Llion Jones, Adam Treat. 🔍 Post-training reimagined & fully open - Pre-trained language models are powerful but incomplete, lacking in safety, alignment, and specialized capabilities. - Tülu 3 introduces a fully open post-training framework, bridging the gap between open-source and leading proprietary models like OpenAI GPT, Anthropic Claude, and Google Gemini. 🛠️ A suite of tools for the community - Open pipelines: fully transparent post-training methodologies, from data curation to evaluation. - Decontaminated datasets: prepped for skills like coding, reasoning, multilingual interactions, and more. - Evaluation framework: simplifies reproducibility and performance comparison for developers. 🌍 Broad impact, scalable innovation - One can mix and match datasets to tailor a model for your specific use case, without losing its core general capabilities. - Scalable across different model sizes (from 8B to 70B parameters), requiring less compute than you’d expect. - Training from scratch still requires money. - Fine-tuning will get much better quickly, requiring to be only GPU-bootstrapped. 🧠 Cutting-edge techniques: - Reinforcement learning with verifiable rewards, skipping traditional reward models. - On-Policy generations: using a model's outputs to refine its preferences. 💡 Why it matters: Tülu 3 is more of an ecosystem than a model. By releasing all code, data, and infrastructure, Ai2 have further empowers nations, researchers, developers, and entrepreneurs to build models that rival closed alternatives while promoting transparency and collaboration. #ai #aiinnovation #aidevelopment #airesearch #aiforgood #artificialintelligence #chatgpt #datascience #futureofai #genai #generativeai #innovation #languagemodels #llama #machinelearning #mltools #mlops #naturalanguageprocessing #nlp #openscience #opensource #opensourceai #openai #posttraining #tülu3

12

1 Comment

A recent totally real and not-at-all made-up survey found that 341 out of 400 data scientists are Bayesian, so we can safely conclude that 85.25% of data scientists are Bayesian. As for me? Well, just off the top of my head, I'm probably a frequentist. But let's review some of my recent analyses and then we can reassess. 😉 Ever feel like you're faking it in the world of data science? You're not alone. 🎭 In data science, it's okay to be uncertain. In fact, understanding and quantifying uncertainty is a crucial part of the job! So next time you're feeling like an imposter, remember: 📊 Even the experts sometimes aren't sure. 📊 It's okay to change your mind when presented with new evidence. 📊 A good data scientist knows when to say "I don't know... yet." What's your favorite data science joke? Share in the comments – bonus points if it makes us question our statistical beliefs! 🤓 #DataScience #Statistics #ImposterSyndrome

2

🗣️James Briggs and Matt Berman on behalf of Pinecone delivered a highly insightful session yesterday, shedding light on the latest advancements in Retrieval-Augmented Generation (RAG) systems. James Briggs in particular, explored new and improved techniques for preparing data for production, highlighting innovative methods like semantic chunking and LLM-based chunking. 🗂️Traditional chunking approaches often require significant manual effort to prepare data before processing. However, these new techniques leverage embedding models to automatically identify semantic structures within the data, streamlining the process and optimising data storage.

11

It's great to revisit dbt after a few years and discover the new Python models they've introduced alongside SQL models. Recently, I tackled a messy set of SQL queries stored in various text files and in my DuckDB history. Cleaning up was essential as the data needed to be processed through a Python script for text manipulation, calls to an Elasticsearch index, and then back into DuckDB. The seamless integration of text manipulation and Elasticsearch calling into "dbt run" was a pleasant surprise. #dbt #Python #SQL #Elasticsearch

12

We are often asked to solve the big problems in healthcare with R and Python and we haven't yet answered the seemingly straight forward questions... I've had this blog https://lnkd.in/eXXnfrD3 open in a tab and finally got around to reading it. Some of these questions feel eerily familiar... "In healthcare, MPs often struggled to get basic prevalence information, whether: - the number of people diagnosed with particular conditions, like silicosis - diagnoses broken down by characteristics, like the number of women with meningitis, or region, like autism in the East of England - or diagnoses for particular (important) groups, like prisoners with mental health conditions. Also, funding is a topic it's surprisingly difficult to get information about, e.g. - funding for particular conditions, like endometriosis, or - funding on social care by local authority - funding at a local level, especially per hospital, which MPs often care about. Following on from this, hospital-level information in general often seems to be poor, e.g.: - how many A&E visits there are per hospital - what waiting times are per hospital. And finally, workforce is a huge one, with topics like: - vacancies - how many current GP vacancies are there? - retention - how many dentists are still working 5 years after qualifying? - skills - how many specialists are there with particular skills, like Parkinson's nurses or motor neurone nurses?... This gets really worrying when you look at the dataset over time. It's immediately clear that MPs often ask the same thing over and over again - yet the information doesn't seem to improve."

19

1 Comment

👉🏼 RDguru: A Conversational Intelligent Agent for Rare Diseases 🤓 Jian Yang 👇🏻 https://lnkd.in/eSCm_MkQ 🔍 Focus on data insights: - 📈 RDguru outperforms ChatGPT in generating authoritative descriptions, highlighting its enhanced accuracy in knowledge Q&A tasks. - 🔍 A novel multi-source fusion diagnostic model shows a significant 5.47% improvement in diagnostic recall compared to PheLR. - 🧬 The use of advanced algorithms facilitates more effective automated phenotype annotation and differential diagnosis in rare diseases. 💡 Main outcomes and implications: - 🏥 RDguru demonstrates the potential to significantly improve clinical decision-making processes in rare disease diagnostics. - 💬 The conversational agent can enhance patient interactions through evidence-based Q&A, leading to better-informed medical consultations. - ⚙️ Integration of multi-source diagnostic strategies could pave the way for the development of more robust AI tools in healthcare settings. 📚 Field significance: - 🚀 This study positions RDguru as a promising solution for addressing the challenges of diagnosing rare diseases, particularly given the limitations of traditional methods. - 🌍 The findings may inspire further research into the implementation of LLMs in clinical environments, promoting wider acceptance and use of AI technologies in healthcare. - 🔒 Ensuring accuracy and reliability in AI outputs is crucial for safeguarding patient safety and enhancing trust in AI applications within medicine. 🗄️: [#RareDiseases #AIAssistedDiagnosis #ClinicalDecisionSupport #MachineLearning #HealthTech #NaturalLanguageProcessing #DiagnosticAccuracy]

1

The Rise of the Data Scientist The Rise of the Data Engineer The Rise of the Analytics Engineer The Rise of the Analytics... Pretendgeneer? Benn Stancil What is next? Data Product Management? Something else? Check it here: https://lnkd.in/eCE3TmP9 #dataengineering #data #orchestration

33

11 Comments

SYNTHETIC DATA On many occasions synthetic data is being discussed as a potential solution. Any sensible conversation about the use of synthetic data starts off by clarifying what we are using that data for: a) Training AI models b) Building Data Pipelines In the case of b) it seems that many do not understand that a hard part of building data pipelines especially on older data sources comes from data not being as expected and that many corner cases must be handled. That makes generating synthetic data for building data pipelines missing a major aspect of what is hard about building data pipelines. Of course it works well for scrambling to ensure privacy and copies to test volume resilience or real-time capabilities. In the case of a) it is a different discussion - still, however, with both pros and cons. There are some obvious invariances which are naturally handled by manipulated data (like rotation, scaling, and translation), while others like representativity is far less obvious - if interested in that part you might like the article linked to below .. #laerdalCPHlife #helpingsavelives #ai https://lnkd.in/dAM5VEUZ

12

This months instalment of my Causal AI series is up on Towards Data Science, this time exploring optimising non-linear treatment effects in Pricing & Promotions (or Marketing): https://lnkd.in/etYraNxt

101

2 Comments

Given my experience with LangChain, building PDF document chunks and generating embeddings through the Palantir Build Pipeline is straightforward. Before you learn this tool, learn the following: Chunks, Embeddings, VectorDB, Similarity Search, Ranking, Prompt, etc in LangChain. Of course, Pyhton is your AK47 Palantir Technologies

10