Peter Smittenaar

𝗙𝗿𝗼𝗺 𝗣𝗿𝗶𝘃𝗮𝗰𝘆 𝘁𝗼 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆: 𝗧𝗮𝗰𝗸𝗹𝗶𝗻𝗴 𝗔𝗜 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲𝘀 𝗮𝘁 𝗡𝗛𝗦 𝗘𝗻𝗴𝗹𝗮𝗻𝗱 I’ve recently written two articles for the NHS England Data Science Website based on some of the really interesting work I’ve been doing! The first: Investigating Annotation Tools for Named Entity Recognition (1). If you like Generative AI, entity extraction and identifying privacy risks - this is for you. The second: The RISE Tool (2) - An Easy Way for Testers and Assurers to Evaluate AI Classifiers. This covers AI evaluation, assurance, image classifiers and explainability. The NHS England Data Science team have a whole host of interesting projects, you can read about them all on our website (3). 1 - https://lnkd.in/dNaUXsgv 2 - https://lnkd.in/dEjp88ra 3 - https://lnkd.in/dYg53A4j

19

It's important to make the most of the data you have to guide efficient compound prioritisation. In next week's webinar, my colleague Ed will discuss multi-parameter optimisation, demonstrating how to build scoring profiles using historical data and analyse the sensitivity of your decisions to the criteria you choose. You can sign up at https://lnkd.in/eHs_cZf9

5

Next posit::conf 2024 rewatch treasure for me: Hannes Mühleisen on duckDB and single node data management. Love the intro, makes you feel a bit less awkward spending so much time thinking about tables :) Why does data management matter to statisticians? Well, I tend to use large scale, very detailed simulations a lot for my work to answer complex questions about design optimization. Nowadays, the computing power to generate this data is not so much the problem any more but rather postprocessing and aggregation. Essentially that's SQL against a massive table of simulation results and I don't want to use a spark cluster for it. Approaches like duckDB have a huge potential for making these workflows simpler, looking forward to exploring this further. Btw. duckDB was at posit::conf last year as well and I completely missed it x)

10

1 Comment

I have seen a lot of posts recently saying "The AI revolution will fail because of Y" and typically "Y" is something that the poster is directly involved in. It may well be the case but the way I see it is there is one simple way in which the AI revolution will fail... If we don't get the entire business involved, and remain too focused on the technicality rather than the business impact. With any new project/advancement/tool answer these simple questions: - Why give a f**k? - Who gives a f**k? - Who's going to f***ing use it? Can't answer these simple questions? Then you have an interesting pursuit, but the business value is questionable. If we fail to learn from previous mistakes and remain insular as a community. It happened with Tech, Data, Big Data, and I am sure there are cases before that. Now you may say "These didn't fail" but ask a random person in a business "How does this company use Z to benefit our partners, clients, or business?" and see what they say. Chances are they won't know or worse won't care as we failed to use these tools to impact their daily lives. Too often as a technical community we focus on the technology rather than the implementation and adoption. If technological advancements are only understood by the technicians, widespread change cannot happen. Spend as much time thinking about integration and deployment in the wider business as you do about the technological advancement and process; and you could well be leading the way in the AI Revolution. #ChangeManagement #AIRevolution #GenerativeAI #TechLeaders #Change

3

2 Comments

Some interesting commentary in this post to consider

4

Fika Bites is back with the inspiring journey of Conor Moag, on how adopting the National Competency Framework right at the start of your career can help you turbo charge it.

6

⚠️ New Viz Alert ⚠️   👀 Let's take a look at my take on this past week's #Makeover Monday.   I haven't seen anyone plot profitability over expectation, and doing so allows us to clearly rank which film gave the greatest return.   But what if the break-even point was different than the industry standard 250%?   And how does the profitability ranking compare to the ranking of other metrics?   ✔ This dashboard answers those questions.   🔗 Link in comments.   🏗→🧠 Build to Learn!   💭🚶‍♀️🚶‍♂️ Follow for more.   #tableau #data #analytics #VizoftheRay Chimdi Nwosu Irene Diomi Harry Beardon

28

5 Comments

🌟 Exciting blog post from NHS England on Reproducible Analytical Pipelines  🥁 Keep banging the drum Sam Hollings 📈 NHS England are sharing their journey on Reproducible Analytical Pipelines (RAP) and how it's enhancing the quality of their data processes. 🔍 What is RAP? RAP aims to automate statistical and analytical processes, making data analysis more efficient, transparent, and robust. This means less manual work and more time for data analysts to focus on what truly matters - delivering insights that improve healthcare services. 🌐 Why It Matters for the Culture of Data Services? Adopting RAP aligns the commitment to enhancing the quality and reliability of data services. It's not just about technology; it's about fostering a culture where continuous improvement and collaboration are at the forefront. 🛠️ Progress NHS England Data teams are integrating RAP across the organisation and It's still early days, but the impact is tangible and transferable. 📈 Looking Ahead They are committed to expanding the use of RAP, enhancing data capabilities, and continuing to share their progress. Hoping to hear more of this type of work. #NHS #DataScience #Innovation #RAP #HealthcareTechnology #Teamwork #Data #Health #Collaboration Link to the source in the comments. Image created by DALLE-3

17

1 Comment

Also in the news - Sam Altman claims that OpenAI IS going to roll out their first AGI in the form of a digital workers (or at least that’s what I inferred from the post) in 2025. https://lnkd.in/gKt3Bwe3

6

2 Comments

Really happy to share that our work on “Learning Metrics that Maximise Power for Accelerated A/B-tests” will appear in the Applied Data Science track of #KDD2024. We address a common problem in industry: how can we minimise type-II errors in online controlled experiments? Can we simply learn a metric combination that directly minimises this quantity? This project started out as “let’s implement the ideas from this great paper by Eugene Kharitonov et al.”, but we ended up proposing some extensions to their seminal work that seem to be quite effective, and are now used across ShareChat. Proud to have this work out there, let us know what you think. Additionally, we were assigned an Area Chair that clearly cared: initial low-confidence reviews prompted them to invite more reviewers (we got 8!), and the AC themselves correctly grilled our rebuttals to point out arguments they deemed insufficient. It’s easy to be positive about peer review when your paper gets accepted, but it seems there are still some unsung heroes out there... Work with Aleksei U.. ACM SIGKDD & Annual KDD Conference

116

9 Comments

Need for diverse sources of #data, like #RWE or real world evidence, was a topic in the #AI panel at Patient-Centered Outcomes Research Institute (PCORI) meeting yesterday. With reports of “model collapse”, we need more data. AI “eating” AI-generated data leads to bad AI. Next week will be speaking at GovAISummit on data lakes and hope to make the case for user-generated data for patient-centered outcomes as part of how we define #quality in #healthcare #datascience.

5

1 Comment

Interesting article https://lnkd.in/eJmPwmrB where the basic something is finding out what the organisation does, even at the senior job levels. About a decade ago, when I was returning to work from extensive maternity leave I got an interview as an analyst at a CCG. These had only just been set up in the NHS and I'd been out if the workforce completely as they were started. The interview went well and I was told my interpretation of a Public Health Spine chart showed my knowledge (I do love a spine chart so I was very enthusiastic to explore it). The feedback in not getting the job was around how I wasn't so familiar about CCGs. The feedback was constructive and supportive but I always wondered how relevant knowing the organisation was, particularly in the NHS where PCTs, Strategic Authorities and countless others had disappeared over the years in restructuring? Applicants give a huge amount of time up to tell you about themselves. To look into the organisation, which will really only give a 2D view to the outside world, does sound like more unnecessary work. We can't freaky know a job from the job description (most have that caveat of any other relevant work!) and we can't really know an organisation from the flashy website. Appointing jobs should be equal. The applicants "sell" themselves and the recruiters "sell" the role and the organisation. The burden over the last 20 years may well have been on the applicant to do both, as this article suggests, so I'm quite pleased it may be changing although articles like this are lamenting the change!

1

In recent years, tech companies have been going through a lot of changes, like optimisations and layoffs, and data teams have felt the impact too—sometimes even more than developers. This is partly because the value that data teams bring isn't always obvious. But this isn’t a new conversation; it’s been around long before the layoffs started. Now, there's something else we need to consider for the future. As data and machine learning become more important, our industry is getting more complicated. This complexity often comes not from the product, business processes, or strict data quality requirements, but from the data teams themselves. Let's talk about data architecture. I've often heard people say, "We can't use PostgreSQL for our data warehouse; we need Snowflake, BigQuery, or Redshift." While this can be true in some cases, it doesn’t really apply to an established company with less than 1TB of raw data. Or take streaming: "We need Kafka to make our reports real-time." While this makes sense, there's rarely a clear answer to questions like, "How often do stakeholders actually check these reports? What’s the downside if they update just once an hour?" Usually, the answer is, "We just need it, just in case." Then there's the use of multiple tools for the same job: "On AWS, we use Athena for SQL access to our data lake. To cut costs, we also use Trino. Plus, we have Redshift as our data warehouse, but we add Clickhouse as a cache to speed up reports." These examples show how we’re making things more complicated than necessary. Complexity here means solving problems that don't really exist. Stakeholders typically don't care what database or data tool you use, as long as the data is reliable and questions are answered quickly. But the cost of maintaining these systems and the staff needed can be huge. Adding streaming where it's not needed means hiring more data engineers with advanced skills beyond basic SQL/Python. It's good if data goes straight to the data warehouse, but often it first goes to a data lake, adding more steps even for simple reports. This ends up costing the business more than necessary. And data architecture is only one example. It applies to our processes, way to collaborate with other teams etc. In the past, when borrowing money was cheap and everyone was hiring, this approach worked. Now, we need to switch to simpler, more efficient ways in our architecture, processes, and team sizes. As data leaders, we should always ask ourselves: "What if we don’t do this? Is there a simpler way to get the same results? Are we solving a real business problem? Does this problem require a complex solution, or are engineers just excited about trying a new tool?" This is my advice for data teams to survive, succeed, and become real partners with stakeholders. What are your thoughts on this? Is it something you're thinking about too? #dataleadership #data #kiss

24

1 Comment

Crunch the numbers and it pays to work in data! A great article in the Business section of The Times by Jane Hamilton that I was privileged to contribute to.  Well worth a read in today’s edition (Friday 29th Nov). Also featured John Eland from STACK Infrastructure which provides several scholarships for data centre trainees. #DataLiteracy #Datacareers #DataEngineering #DataEthics #DataAnalytics https://lnkd.in/eddEAz-3

65

2 Comments

Directed acyclic graphs are fantastic for identifying biases. Why? We want to block all backdoor paths for causal inference. Drawing out the relationships shows all the pathways. It also shows which ones are blocked, and those that need to blocked. We can use this to assess certain biases. For example: selection bias. However it doesn't only apply to selection bias. Confounding, collider bias and more can be identified with a DAG. Drawing a DAG can help prevent a TON of biases. #CausallyCurious #RealWorldEvidence #CausalInference #RealWorldData

22

1 Comment

Over the holiday weekend I came across an insightful article [https://lnkd.in/eeQ6pNAT] by Julian Flaks of EQengineered on the importance of providing due consideration to data quality as data-driven software is developed. It's great to see more thought leaders in our industry emphasizing this critical aspect of data management. In my recent article, I explored similar themes, drawing parallels between data and a well-maintained car, highlighting the necessity of regular maintenance for optimal performance. If you're interested in diving deeper into how proactive data care can drive efficiency and reliability, check out my piece here: https://lnkd.in/eE6Nr4Hy Kudos to Julian for contributing to this important conversation. Let's continue to prioritize data health and drive our industry forward! 🚀 #DataManagement #DataQuality #PreventativeMaintenance #DataCare #TechTrends

4