Want to truly understand how AI learns to make better decisions?
Start with the concept of exploration vs. exploitation and dive into one of the most basic, but super important, ideas in Reinforcement Learning: Multi-Armed Bandits.
It's a simpler start to understand the basics behind. And it might change how you see AI.
→https://towardsdatascience.com/simple-guide-to-multi-armed-bandits-a-key-concept-before-reinforcement-learning/

Why normalize databases?
Yesterday, my tutoring student asked me why databases need to be normalized at all. She said: “Wouldn’t it be easier to just have one big table with all the information?”

It’s a common first question when learning about relational databases.
At first, one big table (e.g. customer name, order date, product name, price) seems easiest.

I told her:
Because that quickly leads to data redundancy, anomalies, and integrity issues when inserting, updating, or deleting records.
Normalization means structuring data into separate, related tables, so that each fact is stored only once. This reduces redundancy & preserves consistency.

Can you remember learning to walk as a baby? You didn’t read a manual. Neither does an AI agent.

Reinforcement Learning (RL) isn’t about knowing the correct answer.
It’s about learning through trial and error, by interacting with an environment & receiving feedback.

That’s how AlphaGo defeated a world champion:
It first learned from expert games. Then it played against itself, millions of times, using RL to get better with each game. That’s how it mastered Go.

Want to really understand how RAG, vector search & chunking work?

Then stop reading theory and build your own chatbot.

This guide shows you how to create a local PDF chatbot using:

LangChain

FAISS (vector DB)

Mistral via Ollama

Python & Streamlit

Step-by-step, from environment setup to deployment. Ideal for learning how Retrieval-Augmented Generation works in practice.

https://medium.com/data-science-collective/rag-in-action-build-your-own-local-pdf-chatbot-as-a-beginner-96c2833869ff

Comment “WANT” if you need the friends link to the article, as you don’t have paid Medium.

Imagine trying two pizzerias and always going back to the one that seemed better. Sounds simple? Maybe.

But what if there’s another one in the city that’s even better – and you never tried it?

That’s what the greedy strategy does. It sticks to what worked best so far.

In Multi-Armed Bandits, a classic problem that helps us understand Reinforcement Learning, there are three common strategies to deal with this dilemma:
Greedy
ε-Greedy
Optimistic Initial Values

https://towardsdatascience.com/simple-guide-to-multi-armed-bandits-a-key-concept-before-reinforcement-learning/

Reinforcement Learning starts with a simple but powerful idea:
Trial & Error. Learning what works.

The Multi-Armed Bandit problem is a first step into this world.
It's not just about slot machines. Iit's about how AI (and humans) learn to choose.

https://towardsdatascience.com/simple-guide-to-multi-armed-bandits-a-key-concept-before-reinforcement-learning/

Do you always go to the same café? Or do you try something new?

That’s the exploration vs. exploitation dilemma: Decision under uncertainty.

Multi-armed bandits model exactly that.

And this dilemma shows up everywhere: Recommender systems, A/B tests, online ads, even in human psychology.

Nobel Prize winner Daniel Kahneman called this one of the most fundamental cognitive patterns.

I explain what it is, why it matters, and how AI systems handle it.

Full article here: https://towardsdatascience.com/simple-guide-to-multi-armed-bandits-a-key-concept-before-reinforcement-learning/

What I learned from an interview with a Harvard postdoc in Biomedicine & AI:
It’s not always about having the right answer.
It’s about asking the questions no one else dares to ask —
even if that requires a lot of patience.

#career #careergrowth #careeradvice
#scientists #research #researcher #AI #ki #computerscience #KI #datascience #datascientist #python #harvard #student

https://medium.com/ai-advances/ai-cancer-cells-a-masters-student-s-journey-5-lessons-from-a-harvard-postdoc-ed9fc50b69d9

I asked a Harvard postdoc a bunch of questions. These are the key takeaways:
The fundamentals still matters: math, statistics, coding & logic.
We need critical thinking & abstract reasoning.
Combining two disciplines leads to exciting projects and connections we couldn’t even imagine before.
A stay abroud is often an experience that changes more than our CV - they leave a mark in our mindset.
Two days in a postdoc's life rarely look exactly the same: Writing papers, training AI models, giving talks at conferences or running project meetings.
Success in research is more like a marathon with long phases without immediate feedback or visible success

-> Grab the full article: https://medium.com/ai-advances/ai-cancer-cells-a-masters-student-s-journey-5-lessons-from-a-harvard-postdoc-ed9fc50b69d9

Write in the comments "friend-link" if you don't have the paid Medium version.

#career #careergrowth #careeradvice
#scientists #research #researcher #AI #ki #computerscience #KI #datascience #datascientist #python #harvard #student

I am seeking a job as a computer science college graduate. I am open to being a machine learning engineer, data scientist, software engineer, or other similar role. I can work near Los Angeles or do remote work. I utilize Python, full-stack development, generative AI, and much more.

Reinforcement Learning doesn’t tell you what’s right.
It only tells you how good your choice was.
No feedback on what to do. Only on how it went.

Example: A multi-armed bandit (like a slot machine with several levers). You don't know which lever is the best - you can only find out by trying it out. Exploring means giving up a known reward (from exploitation) — in hopes of finding a better one.

This balance between exploration and exploitation is the central dilemma in reinforcement learning.

A simple strategy is ε-greedy:
→ In 90% of cases you take the best known action
→ In 10% of cases, you try a different one by chance

In simulations, ε-greedy methods perform better in the long term than pure greed (always take the supposedly best) - because they master the “explore-exploit trade-off”.

What does a baby learning to walk have in common with AlphaGo’s Move 37?

Both learn by doing — not by being told.

That’s the essence of Reinforcement Learning.

In my latest article, I explain Q-learning with a bit Python and the world’s simplest game: Tic Tac Toe.

-> No neural nets.
-> Just some simple states, actions, rewards.

The result? A learning agent in under 100 lines of code.

Perfect if you are curious about how RL really works, before diving into more complex projects.

Concepts covered:
ε-greedy policy
Reward shaping
Value estimation
Exploration vs. exploitation

Read the full article on Towards Data Science → https://towardsdatascience.com/reinforcement-learning-made-simple-build-a-q-learning-agent-in-python/

What does your Spotify account say about your music taste?
I was curious – so I built a little app to find out. Using Python, Streamlit and Plotly, I visualised my top tracks, favourite genres and the release years of the songs I listen to most.

What you need:
Access to the Spotify Web API
A bit of Pandas for analysis
A few lines of code with Plotly
Streamlit to build and run the app

→ The result? A personal, interactive music year in review

I put together a step-by-step guide (beginner-friendly) including code, screenshots, and how to deploy the app on Streamlit Cloud.

Check out the full article here: bit.ly/3SF3VTw

#python #programming #technology #streamlit #spotify #Datavisualization #datascience #datascientist #opensource #api

Friends-Link: https://medium.com/data-science-collective/visualize-your-spotify-data-with-python-and-streamlit-4d4c46bbd39e?sk=12cb785242472daf6d77d024a90ad57c

Stop by and see our CEO @c2main present on using #Linux container groups to improve #PostgreSQL system performance and automation at #PGConfDev in #Montreal tomorrow, May 14th. He'd be happy to chat with you about questions and thoughts relating to #Postgres development, research, and projects!

You see a math formula you don’t immediately understand.

Your instinct? Stop reading.

Don’t.

That’s exactly what I told myself when I started reading Probabilistic Machine Learning – An Introduction by Kevin P. Murphy.

And it was absolutely worth it.

In my last article on Towards Data Science, I took a probabilistic view of supervised, unsupervised & reinforcement learning https://towardsdatascience.com/beyond-glorified-curve-fitting-exploring-the-probabilistic-foundations-of-machine-learning/

Huh... neat!

RE: https://bsky.app/profile/jathansadowski.com/post/3ln23np3vxk2r

the state of california is hiring a senior data scientist

https://innovation.ca.gov/join-us/senior-data-scientist/

Imagine: Text corporas, curated by #libraries, whether #OpenAccess or licensed, ready to use out of the #IDE of a #DataScientist in a secure environment. No need to download lots of data, no need to worry about #AIBots. Here it is:
https://blog.oceanprotocol.com/free-compute-to-data-with-ocean-nodes-vscode-extension-1f8385cb077c

We already delivered a Proof-Of-Concept for @CrossAsia and will continue...

Day 4 of the 5-day Generative AI intensive course by Kaggle and Google covered, among other things, Search Grounding:

With this technique, you can enrich generative AI models with up-to-date and reliable information from the internet — in this case by connecting them to Google Search.

That means the model no longer relies solely on its training data but also integrates information retrieved from search results.

So what's the difference to RAG (Retrieval-Augmented Generation)?

There’s no need to build and manage your own retrieval system (like a vector database), since the additional data is fetched via an API from a search engine.

Here is the link to the corresponding Kaggle notebook: https://www.kaggle.com/code/markishere/day-4-google-search-grounding

What are embeddings?

Embeddings are numerical representations of things like words, sentences, images – or even customer profiles. They help models understand meaning and similarity between those objects.

Example:
The word "king" might be turned into a vector like [0.8, 0.1, …],
while "queen" becomes [0.81, 0.12, …]
→ Because these vectors are close, the words are semantically similar.

Embeddings are the backbone of:
Similarity searches
Recommendation engines
Semantic search
or RAGs (Retrieval-Augmented Generation)

And where are all these vectors stored? In vector databases, which let you search through embeddings lightning-fast.

This was the focus of Day 2 of the 5-Day Kaggle Generative AI Challenge: Embeddings & Vector Stores.