Alexander Molak

Machine Learning, NLP, Causal Inference, Probabilistic Modeling, AI Strategy
Hero Image
Image

About Me

Crossing the boundaries

I am a Machine Learning Engineer and Researcher at Ironscales and an independent Machine Learning Researcher at TensorCell.

I am specialized in natural language processing (NLP), causal inference and probabilistic modeling.

My academic background is in philosophy of language and experimental psychology. Before starting my career in data science, I used to work as a music producer and mixing engineer.

I currently work on a book on Causal Infernece and Discovery in Python (availability H1 2023).

Learn more

Talks & Workshops

Sharing knowledge

I am extremely grateful to all the people who shared their knowledge and experience with me. It's very important to me to give back what I got from others. That's why I love to share my knowledge and experience with others.

Interested in NLP, causal or probabilistic modeling? Join me on one of the upcoming events!

Interested in organizing a training session for your company? Click here!

-

Upcoming events

TBA

--

Selected past events

Causal Inference in Python: Estimating Causal Effects (Keynote)

KI Fabrigk Konferenz, Ingolstadt (Jul 1, 2022)

Practical Causal Discovery in Python

Data Science Summit ML Edition, Warsaw (Jun 22, 2022)

Practical graph neural networks in Python with TensorFlow and Spektral (Workshop)

PyConDE & PyData Berlin 2022 (Apr 13, 2022)

Causality: An Introduction

PyData Hamburg (Mar 29, 2022)

Causal Disovery in Python

GHOST Day Applied ML Conf 2022 (Mar 24, 2022)

Causal Inference in Python: An Introduction

Data Science Summit (Dec 3, 2021)

What should I buy next? How to leverage word embeddings to build an efficient recommender system

PyData Tel Aviv 2021 (Nov 10, 2021)

Modeling aleatoric and epistemic uncertainty using Tensorflow and Tensorflow Probability

PyData Global 2021 (Oct 30, 2021)

Uncertainty? Hands-on Bayesian neural networks with Tensorflow and Tensorflow Probability (Workshop)

NLP & AI Day 2021 (Oct 26, 2021)



Training sessions for your team

Corporate AI Training



lespire.io

Interested in running an effective training session for your team?

Check our programs at lespire.io!

Learn more

Medium Blog

Articles

Blog

Causal Kung-Fu in Python: 3 basic techniques to jump-start your Causal Inference journey tonight

Learn 3 techniques for causal effect identification and implement them in Python without losing months, weeks or days for research

Aleksander Molak

Causal inference, Python

27-Sep-2022

Read More
Blog

Three amazing data science books to read in 2023 (if you won’t manage in 2022)

…and you can read them for free if you want! ❤️

Aleksander Molak

Probabilistic modeling, graph neural networks

04-Feb-2022

Kevin Murphy's Probabilistic Machine Learning: An Introduction, Bayesian Modeling and Computation in Python by Osvaldo Martin and colleagues and Deep Learning on Graphs. What are they about and what to expect inside?

Read More
Blog

Modeling uncertainty in neural networks with TensorFlow Probability - Part 4

Aleksander Molak

Probabilistic modeling, Bayesian neural networks, TensorFlow Probability

26-Nov-2021

Part 4: Going fully probabilistic

Read More
Blog

Modeling uncertainty in neural networks with TensorFlow Probability - Part 3

Aleksander Molak

Probabilistic modeling, Bayesian neural networks, TensorFlow Probability

19-Nov-2021

Part 3: Epistemic uncertainty

Read More
Blog

Modeling uncertainty in neural networks with TensorFlow Probability - Part 2

Aleksander Molak

Probabilistic modeling, Bayesian neural networks, TensorFlow Probability

12-Nov-2021

Part 2: Aleatoric uncertainty

Read More
Blog

Modeling uncertainty in neural networks with TensorFlow Probability - Part 1

Aleksander Molak

Probabilistic modeling, Bayesian neural networks, TensorFlow Probability

03-Nov-2021

Part 1: An Introduction

Read More

Sunday AI Papers

Latest posts

Blog

✨ Are 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 𝗿𝗼𝗯𝘂𝘀𝘁 to 𝘀𝗽𝘂𝗿𝗶𝗼𝘂𝘀 𝗰𝗼𝗿𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀?

Aleksander Molak

Transformers, spurious correlations, invariance

20-Mar-2022

Last Thursday researchers from University of Wisconsin-Madison released a brand new paper analyzing the robustness of ViT model to spurious correlations.

Read More
Blog

✨ Scaling 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 to thousands of layers? 😯

Aleksander Molak

NLP, Transformers, scaling

06-Mar-2022

Last Tuesday researchers form Microsoft Research released a new paper introducing a new method that allows to build 𝗲𝘅𝘁𝗿𝗲𝗺𝗲𝗹𝘆 𝗱𝗲𝗲𝗽 Transformer models.

Read More
Blog

✨ Making 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 4-𝟮𝟭𝘅 𝗳𝗮𝘀𝘁𝗲𝗿, but 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗾𝘂𝗮𝗹𝗶𝘁𝘆 𝗱𝗿𝗼𝗽? Check this! 💥

Aleksander Molak

NLP, Transformers, attention

27-Feb-2022

Last Monday researchers form Cornell University and Google Brain released a new paper presenting 𝗙𝗟𝗔𝗦𝗛 - a novel efficient modification of Transformer architecture.

Read More
Blog

Have you ever thought about 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 your 𝗱𝗮𝘁𝗮 𝗱𝘂𝗿𝗶𝗻𝗴 the 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴? 😯

Aleksander Molak

Deep Learning, training, schedules

20-Feb-2022

On Thursday, Leslie N. Smith of U.S. Naval Research Laboratory released his new paper on general cyclical training. Sound familiar? Maybe, but don't get misled!

Read More
Blog

𝗘𝗻𝗱-𝘁𝗼-𝗲𝗻𝗱 Bayesian 𝗰𝗮𝘂𝘀𝗮𝗹 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 with theoretical guarantees? Yes! 🤯

Aleksander Molak

Causality, Bayesian methods

13-Feb-2022

On the first Friday of February, researchers from Microsoft Research, University of Cambridge, University of Massachusetts Amherst and G-Research released a paper describing a novel method for end-to-end causal process.

Read More
Blog

Commonsense 𝗰𝗮𝘂𝘀𝗮𝗹𝗶𝘁𝘆 using 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀?

Aleksander Molak

NLP, causality, Transformers

06-Feb-2022

Last Monday, researchers from University of Pennsylvania released a paper proposing a new framework to perform 𝗰𝗼𝗺𝗺𝗼𝗻𝘀𝗲𝗻𝘀𝗲 𝗰𝗮𝘂𝘀𝗮𝗹𝗶𝘁𝘆 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 (𝗖𝗖𝗥).

Read More
Blog

Can 𝗰𝗮𝘂𝘀𝗮𝗹𝗶𝘁𝘆 help in making 𝗰𝗼𝗻𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝘁𝗲𝘅𝘁 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 more 𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗹𝗲 and 𝗹𝗲𝘀𝘀 𝗯𝗶𝗮𝘀𝗲𝗱?

Aleksander Molak

NLP, causality, VAE

30-Jan-2022

Last Saturday, researchers from UC San Diego and Amazon AI released a paper describing a novel approach to conditional text generation that leverages causal inference principles to mitigate the effects of spurious correlations.

Read More
Blog

More 𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰𝗮𝗹𝗹𝘆 𝘀𝗲𝗻𝘀𝗶𝘁𝗶𝘃𝗲 contrastive 𝘀𝗲𝗻𝘁𝗲𝗻𝗰𝗲 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀?

Aleksander Molak

NLP, Transformers, contrastive learning

23-Jan-2022

Last Wednesday, researchers from The University of Hong Kong, National University of Defense Technology and SenseTime 商汤科技 released a paper proposing a new contrastive sentence embedding framework called 𝗦𝗡𝗖𝗦𝗘.

Read More
Blog

𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 with efficient 𝘂𝗻𝗰𝗲𝗿𝘁𝗮𝗶𝗻𝘁𝘆 𝗲𝘀𝘁𝗶𝗺𝗮𝘁𝗶𝗼𝗻? 😯 Voilà! 🎉

Aleksander Molak

NLP, Transformers, uncertainty, Bayesian, attention

09-Jan-2022

On the last Monday of December, researchers from University of Amsterdam and Amazon released a paper introducing a novel uncertainty estimation method for Transformers.

Read More
Blog

Is causal 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗲𝗻𝗰𝗼𝗱𝗲𝗱 in the 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲? Making reinforcement learning agents more robust by asking them to 𝗲𝘅𝗽𝗹𝗮𝗶𝗻 their 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀.

Aleksander Molak

RL, NLP, causality, generalizability

12-Dec-2021

Last Wednesday, researchers from DeepMind released a paper describing a novel approach to RL-agent training that makes the agents more robust. Agents were able to learn more generalizable abstractions thanks to... explaining their decisions.

Read More
Blog

𝗚𝗣𝗨-𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗲𝗱 gradient-based 𝗰𝗮𝘂𝘀𝗮𝗹 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 in Python? 𝗛𝗲𝗿𝗲 𝘆𝗼𝘂 𝗴𝗼!

Aleksander Molak

Causality, causal discovery

05-Dec-2021

Last Tuesday, researchers from Huawei Noah's Ark Lab and University of Toronto released a new causal discovery package and an accompanying paper. It brings some really cool features to the table!

Read More
Blog

Can one learn a 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 of a 𝗰𝗮𝘂𝘀𝗮𝗹 𝗴𝗿𝗮𝗽𝗵 with 𝗹𝗮𝘁𝗲𝗻𝘁 𝘃𝗮𝗿𝗶𝗮𝗯𝗹𝗲𝘀?

Aleksander Molak

Causality, causal discovery

28-Nov-2021

Last Sunday, researchers from University of Chicago and Carnegie Mellon University released a paper proposing a novel method of discovering a causal graph with latent variables, but the problem is hard.

Read More
Blog

𝗗𝗶𝘀𝘁𝗶𝗹𝗹𝗶𝗻𝗴 𝗕𝗘𝗥𝗧 better? Get 𝗳𝗮𝘀𝘁𝗲𝗿 and more 𝗿𝗼𝗯𝘂𝘀𝘁!

Aleksander Molak

NLP, Transformers, distillation, compression

21-Nov-2021

Last Thursday, researchers from Intel Labs and University of California, Santa Barbara proposed a new approach to model distillation. The proposed architecture achieves very good trade-off between 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝘁𝗶𝗺𝗲 𝗿𝗲𝗱𝘂𝗰𝘁𝗶𝗼𝗻 and 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆.

Read More
Blog

Contrastive loss 𝗕𝗘𝗥𝗧 pre-training for 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗱 𝗿𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝗮𝘁𝗶𝗼𝗻𝘀? Yes, please!

Aleksander Molak

NLP, Transformers, contrastive loss, pre-training

14-Nov-2021

Last Tuesday, researchers from University of Cambridge, Amazon Web Services (AWS) AI and Monash University released a paper introducing a new pre-training approach leveraging contrastive loss scheme.

Read More
Blog

Ok, computer, 𝗽𝗿𝗲𝗽𝗿𝗼𝗰𝗲𝘀𝘀 my 𝘀𝘁𝗿𝗶𝗻𝗴𝘀! 🤖

Aleksander Molak

NLP, strings, preprocessing

07-Nov-2021

Last Thursday researchers from Eindhoven University of Technology released a paper describing a framework for automated string preprocessing and encoding. The framework leverages probabilistic type inference among other interesting components.

Read More