True Positive Inc.

"This book cleared up a lot of conceptual confusion for me about how Machine Learning actually works—it is a gem of clarity. The worked examples and notebook applications gave me a solid starting point for exploration. Even if you are not planning a career in machine learning applications, this is a solid foundation for thinking about the capabilities of these unique new tools.” —Vint Cerf, Internet pioneer and Turing Award recipient

"The book is a good start for anyone new to language modeling who aspires to improve on state of the art." —Tomáš Mikolov, the author of word2vec and FastText

Author of The Hundred-Page Language Models Book

“If you intend to use machine learning to solve business problems at scale, I'm delighted you got your hands on this book.”

—Cassie Kozyrkov, Chief Decision Scientist at Google

“Foundational work about the reality of building machine learning models in production. Comes at the right time when companies start to see through the AI hype, realizing it takes a conscious engineering effort and best practices to make machine learning work. Another great book from Andriy!”

—Karolis Urbonas, Head of Machine Learning and Science at Amazon

Released on Amazon, B&N, and Leanpub. Drafts are available on http://www.mlebook.com.

Author of the Machine Learning Engineering book

Peter Norvig, Research Director at Google, co-author of AIMA, the most popular AI textbook in the world: "Burkov has undertaken a very useful but impossibly hard task in reducing all of machine learning to 100 pages. He succeeds well in choosing the topics — both theory and practice — that will be useful to practitioners, and for the reader who understands that this is the first 100 (or actually 150) pages you will read, not the last, provides a solid introduction to the field."

Aurélien Géron, Senior AI Engineer, author of the bestseller Hands-On Machine Learning with Scikit-Learn and TensorFlow: "The breadth of topics the book covers is amazing for just 100 pages (plus few bonus pages!). Burkov doesn't hesitate to go into the math equations: that's one thing that short books usually drop. I really liked how the author explains the core concepts in just a few words. The book can be very useful for newcomers in the field, as well as for old-timers who can gain from such a broad view of the field."

CEO and Author of The Hundred-Page Machine Learning Book

MindsDB

I'm happy to share that I'm joining MindsDB as an Advisor.

MindsDB is focused on fascinating and, in my view, crucial aspect of applied AI: seamlessly integrating machine learning capabilities directly into the data layer.

They empower developers to leverage existing databases as predictive engines, using the familiar language of SQL. This dramatically lowers the barrier to building AI-powered applications, abstracting away much of the traditional MLOps complexity.

As you most likely know if you read this post, making advanced technology accessible and understandable has always been a core theme in my work. The approach Jorge and the team have taken – bringing AI to where the data lives and to the tools developers already master – resonates deeply with me.

I see significant potential in the product and the team's vision to democratize AI adoption within the broader software engineering community.

I will mainly contribute by:

- Helping in defining product features, making sure MindsDB continues to solve real-world developer problems with elegance and power.
- Hosting webinars designed to clearly demonstrate the simplicity and versatility of MindsDB for software engineers.

We'll explore practical applications and, of course, I'll be diving into how it all works "behind the scenes" to provide that deeper understanding.

I believe MindsDB is on a compelling trajectory, and I look forward to being part of their journey and engaging with the community.

Advisor

SwarmOne

SwarmOne is a machine learning platform that automates everything: from hardware choice to training distribution strategy. What's especially cool is that it's using machine learning to make decisions based on how various training projects performed in the past in terms of memory consumption, speed, and cost.

What the user needs to provide is only the model they want to fine-tune (all Hugging Face Hub models are supported), the training and validation datasets, and their preferences in terms of training cost vs. speed. SwarmOne's AI will do the rest.

SwarmOne has an API compatible with high-level APIs of Hugging Face, PyTorch, and TensorFlow, so all the user's training pipeline will remain functional with minimal changes as the two screenshots demonstrate.

In my strong opinion, making data scientists think about GPU models, their specs and costs, and training distribution strategies is a waste of their time. They should work on the business problem they are trying to solve, the dataset and the labeling architecture that would work the best, and the hyperparameters they want to fine-tune. All the rest must be decided automatically behind the scenes.

This is exactly what SwarmOne does. You can use it as a standalone service or you can install it in your corporate environment to keep your data secure and let SwarmOne optimize the use of your hardware.

I'm really excited to join SwarmOne at this precise moment, where there are a lot of GPUs, a lot of models, and a lot of users, but there's a huge disconnect between them.

Reach out to me if you would like to have a demo.

Akinn Group

Akinn is working on creating an AI-based investment banking platform. As it turns out, investment banking is a highly manual process. By leveraging modern advancements in machine learning and natural language processing, there's an opportunity to significantly scale the process, and I'm here to guide them and cut through the hype.

Advisor to the Board

TalentNeuron

Leading the team of Machine Learning developers digging through the gigabytes of talent marketplace data.

Tools and techniques used: machine learning (deep/shallow, LLMs, classification, generative AI, clustering, topic modeling), regular expressions, Python, MySQL, scikit-learn, PyTorch.

Machine Learning Lead

Gartner

Leading a team working on "hacking" text and other unstructured or semi-structured data to extract useful information. Problems successfully solved by me and my team include:

— conversational user interfaces (chatbots);
— parsing of free-form résumé to detect and extract candidate name, employers, skills, certifications, employment history;
— parsing of job descriptions to detect and extract job title, salary, employer name, assign a standard occupation code;
— automated noisy and multi-language data normalization, cleanup, deduplication;
— novel data detection;
— language detection / text segmentation by language;
— sequence labeling (queries, addresses, salaries);
— machine translation;
— Web-data analysis.

Tools and techniques used: machine learning (deep/shallow, classification, clustering, topic modeling, structured output prediction), regular expressions, Java, Python, MySQL, scikit-learn, Keras.

Director of Data Science — Machine Learning Team Leader

Wanted Technologies

Turning terabytes of job announcements from online job boards into structured knowledge: extraction and classification of skills, standardization of job titles, classification of job announcements by an occupational classification, extraction/cleanup of structured data (company names, proper names, salaries, skills, etc), extraction and standardization of postal addresses, fuzzy duplicate identification (announcements, skills).

Tools: linux shell scripting, MySQL, Sphinx, Python (scikit-learn, NLTK), Java.

Data Analyst

Fujitsu Canada

Computer scientist with the focus on applying machine learning and natural language processing in the context of big data; technology intelligence lead.

Technologies: R, Python, Hadoop, HBase.

Applied Scientist — Fujitsu Innovation Center

Working on research contracts for the Defence Research and Development Canada (DRDC).

Research projects mostly turned around the problems of fusion of sensor data and intelligent software assistants.

Technologies: Scilab.

Scientific Consultant

Université Laval

Research in multiagent systems resulted in multiple publications in top international conferences and journals (AAMAS, AAAI, Journal of Algorithms).

Student Researcher

Read my just released 📖 The Hundred-Page Language Models Book: https://thelmbook.com

Read my bestselling 📖 The Hundred-Page Machine Learning Book: https://themlbook.com

Read 📖 Machine Learning Engineering: https://mlebook.com

Subscribe to my weekly ✉️ Artificial Intelligence newsletter: https://www.linkedin.com/newsletters/artificial-intelligence-6598352935271358464/ and to my weekly ✉️ Data Science newsletter: https://www.linkedin.com/newsletters/7102511020270608384/

About me:

Ph.D. in Artificial Intelligence, passionate about data, fluent in English, French, and Russian. Solid scientific programming and team leadership skills, with over 20 years of experience working on various computing projects, including several of my own startups.

More than 15 years of hands-on experience in automated data analysis, machine learning, and natural language processing.

Trained a Transformer from scratch and fine-tuned pretrained transformers for various tasks. Built a robot that crawls the internet, finds websites with business-critical information, and retrieves updated information periodically. Developed an enterprise chatbot that doesn’t hallucinate.

Expert in Python and Java with several years of daily design and development experience in big data contexts.

Specialties: machine learning, natural language processing, conversational interfaces (chatbots), information retrieval.

Andriy Burkov

Please stop posting about TOON. Just because you never heard of CSV doesn't mean others haven't either.

Now that you've just learned about CSV, welcome to computing. The next step is this magical thing that allows you to write text and the computer does stuff, called Terminal.

PhD in AI, author of 📖 The Hundred-Page Language Models Book and 📖 The Hundred-Page Machine Learning Book

This paper shows a 27-million parameter model solving very difficult reasoning puzzles—like complex Sudoku and maze problems—using only 1,000 training examples, while current large language models get 0% accuracy on the same tasks.

Instead of having the model generate chains of intermediate reasoning steps as text tokens (like "first check row 1, then check column 2..."), this approach does all reasoning internally by running two coupled recurrent networks that repeatedly refine their hidden states until reaching a solution—one network updates its state every timestep while the other updates only every T timesteps and resets the first network each time it updates.

The training uses a one-step gradient that only backpropagates through the final states rather than through all timesteps, which keeps memory usage constant and makes training stable even when the model runs hundreds of iterative computation steps per problem.

Read it on ChapterPal by talking to it: https://lnkd.in/emT8P5yZ

Download a PDF: https://lnkd.in/etDRGA9B

This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning, traditional tree-based methods (XGBoost, CatBoost, random forests, etc) have dominated tabular data—the most common data format in real-world applications—for two decades. Deep learning conquered images, text, and games, but spreadsheets remained stubbornly resistant.

This paper's (published in Nature by the way) main contribution is a foundation model that finally beats tree-based methods convincingly on small-to-medium datasets, and does so very fast. TabPFN in 2.8 seconds outperforms CatBoost tuned for 4 hours—a 5,000× speedup. That's not incremental; it's a different regime entirely.

The training approach is also fundamentally different. GPT trains on internet text; CLIP trains on image-caption pairs. TabPFN trains on entirely synthetic data—over 100 million artificial datasets generated from causal graphs.

TabPFN generates training data by randomly constructing directed acyclic graphs where each edge applies a random transformation (using neural networks, decision trees, discretization, or noise), then pushes random noise through the root nodes and lets it propagate through the graph—the intermediate values at various nodes become features, one becomes the target, and post-processing adds realistic messiness like missing values and outliers. By training on millions of these synthetic datasets with very different structures, the model learns general prediction strategies without ever seeing real data.

The inference mechanism is also unusual. Rather than finetuning or prompting, TabPFN performs both "training" and prediction in a single forward pass. You feed it your labeled training data and unlabeled test points together, and it outputs predictions immediately. There's no gradient descent at inference time—the model has learned how to learn from examples during pretraining.

The architecture respects tabular structure with two-way attention (across features within a row, then across samples within a column), unlike standard transformers that treat everything as a flat sequence.

So, the transformer has basically learned to do supervised learning.

Talk to the paper on ChapterPal: https://lnkd.in/e_rdrePu

Download the PDF: https://lnkd.in/ekrDRXZY

Frontier LLMs like GPT-o1 or DeepSeek-R1 have achieved much stronger skill by "reasoning" through problems step-by-step, but their internal design is either unknown to the public or known only on the surface.

If someone wants to actually train a skilled reasoning model, a lot of details would be missing and would have to be reinvented, resulting in wasted time and money.

This paper provides the first comprehensive roadmap that breaks these complex systems into understandable, modular components. The authors provide a unified framework showing how reasoning structures, search strategies, and training methods fit together—plus a working implementation called x1 that lets you experiment with these concepts directly.

Read interactively: https://lnkd.in/eQxCkXRd

Download a PDF: https://lnkd.in/eC_miwAH

In this remarkable paper, a team at DeepMind and Terence Tao addressed a fundamental challenge in mathematics using an AI system called AlphaEvolve. The challenge is the following:

Many practical math problems ask "what's the best possible arrangement?"—like packing the most circles in a square or finding the largest set of points with no isosceles triangles—and answering these requires actually finding specific examples from search spaces so large that systematically exploring them by hand is impractical.

AlphaEvolve doesn't have the LLM directly generate candidate solutions (which would require millions of expensive LLM calls to explore the search space)—instead, it evolves Python programs that search for solutions, so one expensive LLM call generates a search heuristic that then runs cheaply for millions of iterations exploring candidate configurations, and the population of these search programs evolves over time with better ones surviving, essentially doing meta-optimization where the thing being optimized is itself an optimizer.

Across 67 problems spanning combinatorics, geometry, and analysis, AlphaEvolve discovered new constructions improving decades-old results (hexagon packing, Heilbronn triangles, sum-product problems) while matching state-of-the-art in dozens of others.

The paper reports every problem they tried including failures where AlphaEvolve couldn't match known bounds, which makes this the most thorough empirical evaluation of AI for mathematical discovery.

Read and ask questions on ChapterPal: https://lnkd.in/ev4WH-bA

Download the PDF: https://lnkd.in/ecvNZaxy

How to lie with charts? Anthropic knows how.

Thanks to Nano Banana, we can see how Opus 4.5 is truly better compared to previous models.

On the left, the graph Anthropic shared. On the right, the graph they should have shared (produced by Nano Banana from the original graph) if they didn't want to deceive the users.

KFC & PIZZA HUT MALAYSIA (QSR Brands (M) Holdings Bhd)

Chief Data Officer

SEEK

I'm responsible for digital media strategy and planning, digital advertising operations, digital analytics, and digital data management across SEEK Asia's businesses in Hong Kong, Malaysia, Singapore, the Philippines, Thailand, and Indonesia.

Regional Head of Digital Marketing, Asia

Accenture

Managing Director, Customer Insights & Growth

ada - analytics · data · advertising

ada Asia is an integrated digital marketing business jointly owned by Axiata Digital Services and Sumitomo Corporation. We combine data science, technology, creative & content, to disrupt marketing for brands and businesses across nine countries in Asia. Our team of over 260 data experts, growth hackers and advertising visionaries leverages extensive experience working with international brands to deliver measurable business results.

Reporting to the CEO, my responsibilities as Head of Data Science & Engineering involves leading a team of 22 data scientists and engineers to develop solutions for consumer data fusion & ontology engineering, consumer preference prediction, consumer action prediction, causal inference, and decision optimization.

Selected Contributions:

	Led the development of market insights products for the airlines, telecommunications, handset manufacturing, and retail industries.
	Led the development of ada’s data management platform, ada Xact, which contains around a hundred attributes on nearly 280 million internet users in nine countries.
	Led the development of ada’s mobile software development kit for capturing in-app events from clients’ mobile apps and a stream-processing backend system to collect in-app events from up to 4 million users.
	Led the development of ada’s campaign performance database and advertising effectiveness models for both online and out of home media.
	Using the co-location events of audiences and graph theoretic methods, we built a set of models to score audiences for their potential social influence.
	Using image feature extraction and regression analysis methods, we built a model to predict the click through rates for display ads produced by our creative agency.

Head of Data Science and Engineering

Axiata Digital

Axiata Group is an Asian telecommunications group with approximately 320 million subscribers and 25,000 employees in ten countries. Axiata Group’s portfolio of services in mobile communications services, tower infrastructure, and digital services generates over RM21.6 billion (USD5.2 billion) in revenues, RM657 million (PAT USD159) in after tax profits and a market capitalization of RM42.3 billion (USD9.4 billion).

In 2012, Axiata established Axiata Digital to capture the rapid growth in Internet-based businesses. By the end of 2016, Axiata Digital has built a portfolio of 29 companies in digital advertising, digital marketplaces, digital financial services, digital entertainment, data science, and over-the-top (OTT) enablement platforms. Axiata Digital companies served over 32 million customers and generated around USD300 million in gross revenues in 2016.

As Head of Analytics, I help Axiata Digital’s clients optimize their core business processes using intelligent automation.

Selected Contributions:

	Developed a set of intelligent automation solutions for marketing and sales management, customer management, and service management.
	Grew signings for analytics services by over 390% in second half of 2016 after I joined Axiata Digital.

Head of Analytics

IBM is the world’s leading cognitive solutions and cloud platform company with annual revenues of over $80 billion.

I apply data science and design thinking to help IBM clients advance their pricing, customer management, and digital advertising capabilities.

Clients Served:

 Indonesia – Indosat Ooredoo
 Malaysia – Axiata Group, Packet One Networks
 Singapore – StarHub
 Thailand – True Corporation
 UAE – Etisalat 

Selected Contributions:

 Developed three solution frameworks in the areas of pricing, customer management, and digital advertising. This includes algorithm design, data modeling, knowledge development, and organizational design.
 Awarded the IBM Global Business Services Service Excellence Award (South East Asia) - 2015 for my work in solution design.

Associate Partner - Global Center of Competence for Telecommunications, Media, and Entertainment

Celcom Axiata Berhad

Celcom Axiata Berhad is the second largest communications services provider in Malaysia.

Reporting to the Chief Marketing Officer, my current role as SVP / Head of Intelligence involves leading a team of 50 analysts that offers market research, customer analytics, campaign optimization, pricing strategy and marketing profitability services to internal clients.

Selected Contributions:

 Using multiple linear regression analysis on campaign response data, we discovered six campaign attributes that lead to increased revenue per subscriber for prepaid bundle cross/upsell campaigns. Apply these insights, we were able to grow prepaid bundle revenue by 50% over a 12 month period.
 We applied conjoint analysis to measure consumer preferences over a variety of prepaid bundle attributes and to design prepaid bundles that both enhanced consumer utility and improved contribution margins.
 Using tariff curves, we were able to precisely simulate the impact of competitive product launches on our customers an confidently decide on the appropriate competitive responses.
 We’ve applied constraint optimization techniques to calculate the right selling price for new product bundles given the desired bundle attributes and levels. As a result, we’ve been able to defend our voice, messaging and data yields in an industry facing falling yields.
 Using cohort analysis, we developed an automated report to analyze the effectiveness of customer acquisition activities along the measures of cohort size, revenue per customer, survival rate, % of active customers, % of time customers are active, and payment delinquency rates.
 We applied design of experiments (DOE) methodology to plan marketing campaigns as market experiments so that the campaign results obtained can be analyzed to yield valid and reliable guidelines for future campaigns. In DOE, we deliberately changed one or campaign variables in order to observe the effect the changes have on campaign responses.

Senior Vice President / Head of Intelligence

Reporting to the Senior Vice President of Customer Value Management, my primary responsibility is to provide quantitative analysis to optimize: 

•	Acquisition quality
•	Tariff migrations
•	Cross / up-sell
•	Retention & churn management
•	Loyalty management
•	Customer lifecycle management
•	Product profitability

This include leading a team that offers the following analytical services to other departments:

1.	Analytical data management
2.	Pre- and post-campaign analysis
3.	Customer segmentation
4.	Predictive modeling and forecasting
5.	Constraint optimization

Selected Contributions:

 Using the Markov Chains technique, we built a churn trigger model for the prepaid base that is up to 20% more accurate than traditional churn models.
 We developed analytical rules for classifying customers according to their lifecycle stage –activation, growth, stability, decline, etc. – allowing us to more appropriately match campaign offers to the customer’s life stage.
 Develop a churn offer matrix that determines the maximum offer amount we can make to each subscriber based on her ARPU, tenure, churn risk and (for a prepaid subscriber) network expiry status while maintaining an expected campaign ROI of >= 200%.
 Using time series decomposition on our daily reloads, we managed to extract the de-seasonalized reload trend and use it to identify which ATL campaign are driving reloads. As a result, we found that Celcom’s daily reloads trended higher on days when it ran entertainment-related ATL campaigns, provided the featured artistes have a large local following.

Vice President, Customer Analytics

Selected Contributions:

 Using Clustering Analysis on top of 750 demographic and behavioral variables, we segmented both the prepaid and postpaid customer bases; allowing us to design segment-specific initiatives that can deliver over RM200 million in incremental revenue.
 Using the Auto-Regressive Integrated Moving Average (ARIMA) method, we developed forecasts for Celcom's daily retail revenue that raised forecast accuracy to over 95%.
 Using the Association Rules method, we discovered 15 product combinations that, when cross-sold, can potentially add over RM40 million per month to Celcom's retail revenue.

Senior Manager, Customer Insights

OMG Asia

With over 10,000 employees in 8 countries, OMG Asia is the largest integrated retail marketing services company in Asia. From in-store media and in-store demonstrations to precision marketing and data analytics, OMG Asia’s products and services aim to help retailers attract shoppers and marketers to build brand loyalty through precise and effective means.

Reporting to the Chief Technology Officer, my primary responsibilities as Senior Manager in OMG Asia include: (a) managing the scope, time, cost, quality, human resources, communications, risk and procurement for all data analytics projects; (b) advising top management on their data analytics strategy; and (c) client consulting on data analytics.

Selected Contributions:

 Lead the development of a new retail marketing data warehouse and standard reports catalog based on Oracle 11g.
 Lead the implementation of three data mining products – shopper segmentation, dormancy prediction and category acquisition – for TOPS Markets, Danone Dumex, Kao, Coca-Cola, Unilever, Johnson & Johnson, Cadbury, Nestle, Mars, Loreal, Colgate-Palmolive and Beiersdorf.
 Developing new data mining products to be offered in 2010 – namely demand forecasting, queue modeling and revenue risk modeling.

Senior Manager, Data Mining

Biztrak Accounting

Biztrak Business Solutions Sdn Bhd is one of Malaysia’s largest financial accounting & business intelligence software company (now part of Queensland-based, ASX-listed Cape Range Limited).

Reporting to the Managing Director, my primary responsibilities as General Manager in Biztrak include: (a) driving top-line growth through product and service innovation; (b) improving productivity through continuous process improvement and human resource development; and (c) providing pre-sales consultation to clients on data warehousing and data mining.

In my earlier position as Senior Manager of Strategic Planning in Biztrak, I was responsible for: (a) market research and industry analysis; (b) market segmentation, targeting and positioning; and (c) formulating and implementing marketing programs.

Selected Contributions:

2008
	Lead the development of a series of functional analytics solutions for marketing, HR, finance and supply chain management based on Microsoft SQL Server BI.
	Formulated data mining strategies for two regional telcos.

2007
	Lead the acquisition of a leading Malaysian business intelligence software company.
	Lead the development of Biztrak Percepta – a front-end solution to Microsoft SQL Server BI.
	Authored a research paper on performance management.

2006
	Formulated and implemented the company’s transformation strategy.
	Rationalized the company’s product portfolio by eliminating non-performing products (57% of the product portfolio).
	Slashed operating expenses by over 26% without impairing operating capabilities.
	Grew sales by 14% despite adverse industry conditions.
	Grew the value-added reseller network by 58% (from 38 to 60 active channel partners).

General Manager

Karensoft Technology Bhd

Karensoft Technology Bhd was one of Malaysia’s leading ERP software company and was listed on the MESDAQ market of Bursa Malaysia.

Reporting to the Executive Chairman/CEO, my role as Manager for Strategic Planning & Business Consulting in Karensoft includes responsibilities such as: (a) market research and industry analysis; (b) developing and implementing the marketing mix; (c) providing pre-sales consultation to clients on business process improvement.

Selected Contributions:

	Co-formulated a growth strategy that delivered revenue growth of 246% in 2003 (as verified by Deloitte Touche Tohmatsu Asia Pacific’s Technology Fast 500 report).
	Delivered 11 public speaking engagements to both domestic and international audiences.
	Authored research reports distributed at analysts/investors briefings. Each report generated widespread interest in the company leading to positive gains in the company’s share price.
	Authored four solution white papers for distribution to clients.
	Provided business process consulting services to 12 clients from the palm oil, timber, rubber glove, automotive, steel, engineering services and semiconductor industries.

Manager, Strategic Planning & Business Consulting

Chief Data Officer at QSR Brands

To be wise and true and subversive and, most importantly, to forge a connection between people.

Disclaimer: The views, thoughts, and opinions expressed in the author's articles, posts, or documents belong solely to the author, and not necessarily to the author’s employer, organization, committee or other group or individual.

taplio

The Battle

Andriy Burkov

George C.

Followers

Posts

Engagements

Likes

Performance ScoreAndriy Burkov's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Andriy Burkov

George C.

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Andriy Burkov's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAndriy Burkov's LinkedIn Performance