Hugo Penedones

Equivariance in Neural Networks: A Free Lunch That Isn't

Tue, 14 Apr 2026 00:00:00 +0000

Some inductive biases encode beliefs about the data: bets that certain hand-crafted features matter, that certain regularities hold. These can be wrong. But some encode theorems about the target function. These cannot be wrong. Rotational equivariance in molecular simulation is one of them.

This post is about why encoding a provably correct theorem into a neural network is harder than it sounds.

What equivariance means

A function $f$ is equivariant with respect to a group $G$ if transforming the input produces a correspondingly transformed output ¹:

Running LLMs on AMD Ryzen AI NPU under Linux — A First

Sun, 08 Mar 2026 00:00:00 +0000

I just got Llama 3.2 1B running at 60 tokens/sec on my nucbox mini PC’s NPU (Neural Processing Unit) (AMD Ryzen AI 9 HX 370) under Ubuntu Linux — something that, as far as I can tell, hasn’t been publicly demonstrated before.

The problem

AMD shipped the XDNA2 NPU in Ryzen AI processors (Strix Point, Strix Halo, Kraken Point) with up to 50 TOPS of AI compute. On Windows, tools like Ryzen AI Software and FastFlowLM make it easy to run LLMs on the NPU. On Linux? Not so much.

The Future of AI is Physical: Simulation is Key

Fri, 07 Nov 2025 00:00:00 +0000

Everyone’s talking about LLMs, but it is already pretty clear what the next big wave will be: AI for the physical world. AI that understands intuitive physics, not only the mechanics of large rigid bodies, which is fundamental for Robotics and autonomous driving, but also AI that understands fluid dynamics, thermodynamics, electromagnetism, plasmas, and even the quantum physics that governs the small scale of atoms and molecules. Every major industry stands to gain, including automotive and transportation, renewable energy (wind, solar, nuclear), pharmaceutical and materials development, you name it!

Ciência 2025: AI for Science Roundtable

Wed, 09 Jul 2025 00:00:00 +0000

Arriving at the Portuguese encounter “Ciência 2025”, for a roundtable discussion on AI for Science and how Inductiva.AI is contributing with a platform that makes numerical simulation and physics datasets generation really easy and affordable.

At Nova School of Business and Economics .

Here is a 1 min flash interview I gave at the event, sharing my thoughts on the role of AI and numerical simulation for science and engineering.

Machine Learning Summer School at University of Porto

Wed, 30 Apr 2025 00:00:00 +0000

Machine Learning Summer School coming up!

Inductiva.AI , in collaboration with the Universidade do Porto and the support of the AptWind doctoral network, is hosting a 5-day programme where we will dive into the state-of-the-art Machine Learning techniques and their impact across different scientific fields. The primary focus will be on Computational Fluid Dynamics (CFD), but it will also explore broader engineering and scientific computing applications.

🗓️ Dates: 19-23 May 2025
📍 Location: University of Porto

DeepMinders at AIHub Lisbon

Thu, 27 Feb 2025 00:00:00 +0000

🇵🇹 DeepMinders Joao Carreira and Zita Marinho, with ex-DeepMinders Wang Ling and Hugo Penedones at AIHub / Unicorn Factory Lisboa in Lisbon.

AI meets Scientific Computing: Transforming Biology and Medicine

Wed, 11 Dec 2024 00:00:00 +0000

I was invited as the keynote speaker at the Dia da Investigação (Research Day) organised by the Faculdade de Medicina da Universidade de Lisboa , held at the Auditório Professor David Ferreira (A52), Edifício Egas Moniz, at 13h.

The talk was titled “AI meets Scientific Computing: Transforming Biology and Medicine” and covered the intersection of artificial intelligence, scientific computing, and their transformative impact on biology and medicine - with a particular focus on how AI is reshaping drug discovery, protein structure prediction, and life sciences research.

Talk at FCUP: Machine Learning and the 2024 Nobel Prizes

Tue, 22 Oct 2024 00:00:00 +0000

Hoje as 17:30 na Faculdade de Ciências da Universidade do Porto (FCUP), uma palestra sobre as contribuições da área de Machine Learning nos prémios Nobel de 2024. Apareçam! :)

(Today at 17:30 at the Faculty of Sciences of the University of Porto, a talk about the contributions of Machine Learning to the 2024 Nobel Prizes. Come join!)

A Data Bottleneck is Holding AI Science Back

Tue, 15 Oct 2024 00:00:00 +0000

“If there were many databases as good as the PDB, I would say, yes, this [prize] probably is just the first of many, but it is kind of a unique database in biology” - David Baker (Nobel prize in Chemistry).

Agreed. That’s why in most other problems in science and engineering, we will need to generate high quality synthetic data using numerical simulators where the rules of Physics are programmed explicitly. You can use them to generate big datasets, with perfect annotations for supervised learning (think: fluid dynamics where you know the velocities at every point in space and time).

2024 Nobel Prize in Chemistry: AlphaFold

Wed, 09 Oct 2024 00:00:00 +0000

Congrats Demis Hassabis and John Jumper! :) 😃

Nobel Prize in Chemistry 2024 winners David Baker, Demis Hassabis and John Jumper. Illustration: Niklas Elmehed / Nobel Prize Outreach

The Royal Swedish Academy of Sciences awarded the 2024 Nobel Prize in Chemistry with one half to David Baker “for computational protein design” and the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction.”

AI for Science Really Means Engineering for Science

Tue, 08 Oct 2024 00:00:00 +0000

Steve Crossan’s article Engineering for Science asks:

“Why did AlphaFold happen at DeepMind rather than (for example) the Broad Institute ? It wasn’t data. Everyone had access to exactly the same data. It wasn’t compute. The compute budget for AlphaFold1 was well within the budget of an academic project. The real reason was that we treated it as an engineering problem as much as a research one. In fact, this was the secret sauce of DeepMind: around ⅓ of the overall headcount was devoted to what we called Research Engineering.”

When Schmidhuber Cites Your RL Paper

Sat, 11 May 2024 00:00:00 +0000

That feeling when Schmidhuber cites your (not well known) RL paper! 😝

Our 2019 NeurIPS paper “Adaptive Temporal-Difference Learning for Policy Evaluation with Per-State Uncertainty Estimates” didn’t catch much attention, but it actually addresses a very fundamental problem in Deep Reinforcement Learning: how can you trust Temporal Difference updates, when you are not in a tabular setting, and instead your function approximator is a neural network?

Forecasted the Future: AlphaGeometry

Wed, 17 Jan 2024 00:00:00 +0000

Forecasted the future 2 days ahead, not bad! Without inside info, despite having worked at DeepMind in the past. :)

Two days ago I gave a talk at the Math department of Faculdade de Ciências e Tecnologia da Universidade NOVA de Lisboa, where in the last slide I made a prediction:

In 2024 we would see a breakthrough in LLMs for Math and it would probably involve:

merging search with LLMs, much like System 1 and System 2 in “Thinking Fast and Slow” and AlphaGo
solving International Math Olympiads problems at a level similar to the best competitors

See a screenshot of my last slides.

AI-Driven Drug Discovery: Why Molecular Dynamics Simulations Matter

Tue, 09 Jan 2024 00:00:00 +0000

Prediction: training data from molecular dynamics simulations will be key.

This is a reaction to the VentureBeat article “AI-driven drug discovery is poised to boom in 2024” .

Background story: I was part of the initial AlphaFold team at DeepMind, back in early 2016. We knew that ideally we would train a deep neural network end-to-end on a large dataset of mappings (amino-acid sequence -> 3d structure), however the data in the Protein Data Bank was not enough to do it “easily”, out-of-the-box. If only we could generate synthetic data at scale! We thought of using molecular dynamics simulations: we would spend a lot of compute time (ok, inside Google) and then distil that knowledge into the neural network, much like AlphaGo did. Unfortunately, even that was out of reach: molecular dynamics simulators could only fold relatively small proteins, and just setting up the infrastructure to do that at scale was hard, even for an elite research organisation. We ended up sticking to the experimental data available, adding extra unsupervised proteomics data (multiple sequence alignments), and tried a lot of tricks to somehow encode some physical priors into the neural network architecture - as well as include some sort of search process, guided by physics potentials.

Artificial General Intelligence is Already Here

Tue, 10 Oct 2023 00:00:00 +0000

I tend to agree with Blaise Aguera y Arcas and Peter Norvig, who argue in their Noema Magazine article “Artificial General Intelligence Is Already Here” that today’s most advanced AI models should already be recognised as the first true examples of AGI.

Their key insight: AI systems used to perform better than humans on some specific narrow tasks (e.g. chess). But they couldn’t do anything else.

Now the situation is reversed: you can still find human experts that give better answers than ChatGPT in each specific subfield, but no single human can give good answers on such a wide range of topics (and communicate in so many different languages) - if that person existed, they would be considered a genius with an unprecedented level of knowledge.

AlphaFold Team Wins the Lasker Award

Thu, 21 Sep 2023 00:00:00 +0000

Congrats to my ex-colleagues from the AlphaFold project. It all started with a small internal Hackathon, and several months of hard work to just get near the state-of-the-art… Impressive how successful it has become! 🧬 🚀

“Never underestimate the power of a small group of committed people to change the world. In fact, it is the only thing that ever has.” - Margaret Mead

Para bom entendedor, meia palavra basta... completa o ChatGPT

Fri, 02 Jun 2023 00:00:00 +0000

Talvez não seja óbvio, mas ser capaz de prever a próxima palavra, pode ir muito além de identificar correlações estatísticas superficiais: imaginem um romance policial em que após centenas de páginas com detalhes intrincados sobre todas as personagens, o detetive diz “o culpado é …”.

O ChatGPT, criado pela OpenAI, é baseado numa grande rede neuronal artificial treinada exatamente para a tarefa de prever a próxima palavra de um texto. Mas para um volume de dados muito superior ao que um humano conseguiria ler durante uma vida! Imaginem: todos os artigos da Wikipédia em todas as línguas, milhares de livros, milhões de páginas web e programas de computador nas mais variadas linguagens de programação. Esta fase de treino força o modelo a comprimir muitíssima informação num número limitado de neurónios artificiais, dando origem a representações numéricas muito boas do significado de palavras e frases. Ao ponto de aprender relações de proximidade visual entre as diferentes cores (ex: framboesa é mais parecido com vermelho do que com verde), sem nunca ver imagens! Aprende até a prever o próximo movimento num jogo de xadrez, pois na internet há muitas listagens de jogos completos. O ChatGPT é depois refinado num regime de aprendizagem por reforço, para dialogar com humanos e dar respostas que sejam consideradas úteis respeitando certas normas sociais.

Programming Principles

Tue, 30 May 2023 00:00:00 +0000

Computation is at the heart of what we do. We write computer programs for running simulations of physical systems, training Machine Learning models, automating tasks or simply testing new ideas. In this blog post we share the programming principles that we try to follow to make fast progress on our goals, while maintaining high standards on the quality of our code.

Principles:

Tracer Bullets
Readability first
Simplicity
DRY: Don’t Repeat Yourself
Don’t write “scripts”
Write Unit Tests
Manage Technical Debt
Don’t Reinvent the Wheel

Tracer Bullets

In The Pragmatic Programmer book, the authors talk about Tracer Bullets in the context of:

AlphaFold: Using AI for scientific discovery

Wed, 15 Jan 2020 00:00:00 +0000

Our Nature paper describing AlphaFold is finally out. :) You can read it at: https://rdcu.be/b0mtx #nature #proteins #structureprediction #deepmind #alphafold #deeplearning

AlphaFold: Using AI for scientific discovery

Adaptive Temporal-Difference Learning for Policy Evaluation with Per-State Uncertainty Estimates

Fri, 06 Sep 2019 00:00:00 +0000

Our paper was accepted at NeurIPS 2019. Yay! :) We use ensembles of neural networks to get (per-state) uncertainty estimates and dynamically switch between Temporal-Difference and Monte Carlo estimates. Now you can do more accurate on-policy evaluation from logs. #reinforcementlearning #neurips

Work done in collaboration with Carlos Riquelme, Damien Vincent, Hartmut Maennel, Timothy Mann, Andre Barreto, Sylvain Gelly and Gergely Neu.

Adaptive Temporal-Difference Learning for Policy Evaluation with Per-State Uncertainty Estimates

My GitHub repositories

Sun, 25 Sep 2016 00:00:00 +0000

I recently made public more of my programming projects from the past. My github profile now includes:

luacnn - Convolutional Neural Network for hand digit recognition using Torch7 and Lua.
metrics - a Torch7 package to compute some metrics, such as area under ROC.
lakeml - some Machine Learning algorithms implemented in C++, e.g. AdaBoost, K-means, EM for (diagonal) Gaussian Mixture Models.
partitracker - a simple Particle Filter implemented in C++ and OpenCV.
sudoku - a Sudoku puzzle solver in pure C.
kdtree - a K-d tree C++ implementation, for Nearest Neighbours search.
fractals - Zoomable Mandelbrot fractal renderings using PyGame.
acm_utils - Data structures and algorithms useful for programming competitions (trie, balanced tree, big ints, suffix arrays, etc.) and other small C or C++ pearls.

Top cited Portuguese Machine Learning researchers

Fri, 25 Sep 2015 00:00:00 +0000

It turns out that there are several very high profile Portuguese researchers in the field of Machine Learning, even though they all seem to live and work abroad. Here are, to the best of my knowledge, the most cited ones, including their current affiliation and the total citations according to Google Scholar (as of Sept. 2015):

Fernando Pereira (Google, USA) citations: 32496
Pedro Domingos (University of Washington, USA) citations: 27248
Manuela Veloso (Carnegie Mellon University, USA) citations: 22405
Nando de Freitas (Oxford University and Google DeepMind, UK) citations: 18439
Jose Principe (University of Florida, USA) citations: 16823
Nuno Vasconcelos (University of California San Diego, USA) citations: 8123

Joined Google DeepMind

Fri, 21 Aug 2015 00:00:00 +0000

I have recently started working as a Research Engineer at Google DeepMind , in London. It’s very exciting to be surrounded by some of the best researchers in Machine Learning in the world, with such an ambitious mission statement and the right conditions to pursue it. I will be working in Deep Learning and Reinforcement Learning real world applications with positive impact in society.

Go ahead and read some of the coolest publications from my colleagues.

Yoshua Bengio's talk in London

Thu, 16 Apr 2015 00:00:00 +0000

Yesterday I attended a talk organized by the London Machine Learning meetup group, where Yoshua Bengio was the invited speaker . Not surprisingly, there were about 200 people attending.

Yoshua reinforced the idea that a lot of the success of learning algorithms for AI tasks comes from incorporating meaningful priors. These should be general enough to hold true in a wide range of applications, but also specific enough to vastly reduce the amount of training data needed to achieve good generalization. This reminded me of a previous post I wrote in this blog, almost 5 years ago!

Contributing to Torch

Tue, 30 Sep 2014 00:00:00 +0000

I recently started playing with the torch library again. Torch7 is now a growing set of packages, managed by luarocks. I really like this approach because it forces torch contributors to make their code more modular and re-usable.

So far, I have done a few very simple contributions to the torch ecosystem:

Extended the matio package, which reads MAT files, to support structs, cell arrays and strings, in addition to loading tensors, which was already implemented.

Microsoft techfest and ML conference

Sun, 09 Mar 2014 00:00:00 +0000

Ten years after my Summer internship at Microsoft as an undergraduate student, I had the opportunity to be back to the Redmond campus. This time the goal was to attend Microsoft Research’s techfest and participate in the Practice of Machine Learning conference. I can’t share much about what I saw there, but I must say I was impressed by the amount of great talks and cool demos. MSR is really a leading research institution and lots of theory and cutting-edge applications keep on emerging, most notably in many Bing products. It’s great to be part of this community. It was also a good occasion to see some old friends, which made my stay even richer. Looking forward to be back!

Inspiring people and their works

Thu, 29 Aug 2013 00:00:00 +0000

Nassim Nicholas Taleb (author, scholar, statistician, trader, philosopher, risk manager) Books: The Black Swan ), Antifragile: Things That Gain from Disorder

Bret Victor (engineer, designer, inventor, dreamer): Talks: Inventing on Principle , The Future of Programming Essay: Learnable Programming

Salvatore Sanfilippo (programmer, I mean rock-star programmer): Software: redis (source code )

Attending SIGIR 2013

Fri, 26 Jul 2013 00:00:00 +0000

Soon I will be in Dublin, Ireland, attending one of the best conferences in Information Retrieval. Looking forward to hearing about the latest research and probably meeting some old friends.

http://www.sigir2013.ie/

The rise of Deep Learning

Sun, 14 Jul 2013 00:00:00 +0000

A couple of years ago, I was doing research involving Convolutional Neural Networks for object classification tasks (see here ). At that point, deep learning was an (re)emergent field, but it hadn’t achieved mainstream yet.

Today, several factors are contributing to the increase of visibility:

Announcements of great results in Kaggle competitions and commercial applications, such as Google’s improved photo search and Microsoft’s fast and robust speech recognition system .
Online courses, such as Geoffrey Hinton’s “Neural Networks for Machine Learning” in Coursera.
Greater variety of software implementations, including many that use the power of GPU parallelism.

What will be the next milestones?

Machine Learning Workshop - Idiap EPFL 2012

Tue, 20 Nov 2012 00:00:00 +0000

Yesterday I attended to the Machine Learning Workshop at Idiap EPFL.

It was a good opportunity to see old friends and colleagues, and listen about their latest research. In general, the quality of the talks was quite good, ranging from very theoretical machine learning (sparse coding, optimization, etc.) to commercial applications of computer vision.
Somewhere in the middle of that spectrum, I also quite liked the talk about learning image local descriptors (BRIEF and LBGM) as a compact and efficient alternative to SIFT or SURF, which are hand-designed, slower and use more bits. There were also applications to speech, face analysis and even remote sensing.

Active Appearance Models

Mon, 12 Nov 2012 00:00:00 +0000

Lately, I have been working with Deformable Models and I am surprised by how well they can work.

Your browser does not support the video tag.

In the video above I am using an Inverse Compositional Active Appearance Model, which was trained with images of myself. It’s specially tuned for my face, but I still find it quite impressive how well it can track my face in realtime!
On the other hand, this model is quite sensitive to lighting conditions and partial occlusions. Training it, is also somehow of an art, because, as opposed to discriminative models, increasing the amount of training data might actually decrease performance. This happens because we use PCA to learn the linear models of shape and texture, which will degrade if data has too much variation or noise.
Still, it’s quite impressive what one can achieve by annotating a few images (about 50, in this case). In addition, as one annotates images, one can start training models that will help us landmark the next ones (in a process of “bootstrapping”, similar to the one in compilers).

The AI set of functions

Fri, 12 Nov 2010 00:00:00 +0000

I recently read an article from Y. Bengio and Y. LeCun named “Scaling Learning Algorithms to AI” . You can also find it as a book chapter in “Large-Scale Kernel Machines"L. Bottou, O. Chapelle, D. DeCoste, J. Weston (eds) MIT Press, 2007.

In some aspects it is an “opinion paper” where the authors advocate for deep learning architectures and their vision of the Machine Learning. However, I think the main message is extremely relevant. I was actually surprised to see how much it agrees with my own opinions.
Here is how I would summarize it:

Tutorial: handwritten digit recognition with convolutional neural networks

Mon, 08 Nov 2010 00:00:00 +0000

I recently added to my webpage a tutorial on how to use torch5 library to train a convolutional neural network for the task of handwritten digit recognition.

Update (February 20, 2026): The code for this tutorial is now available at https://github.com/hpenedones/luacnn and uses torch7. This tutorial was reviewed by Claude Code as of 2026.

NYC Machine Learning Symposium 2010

Sat, 23 Oct 2010 00:00:00 +0000

The event took place yesterday at the New York Academy of Sciences, a building right next to the World Trade Center. The views from the 40th floor were breathtaking:

The names of the participants in the room was no less impressive, (by no special order): Corinna Cortes (Google), Rob Schapire and David Blei (Princeton University), John Langford and Alex Smola (Yahoo), Yann LeCun (NYU), Sanjoy Dasgupta (Univ. California), Michael Collins (MIT), Patrick Haffner (AT&T), among many others. I particularly liked to see the latest developments in LeCun’s group, including a demo by Benoit Corda and Clément Farabet on speeding-up Convolutional Neural Networks with GPUs and FPGAs. Alex Kulezka and Ben Taskar had a nice work on “Structured Determinantal Point Processes” , which can be seen as a probabilistic model with a bias towards diversity of the hidden structures. Mathew Hoffman (with D. Blei and F. Bach) used stochastic gradient descent (widely used among neural network community) for online training of topic models . Sean Gerrish and D. Blei actually had a funny application of topic models to the prediction of votes by Senators !I was also happy to see that there is some Machine Learning being applied to the problem of sustainability and the environment. Gregory Moore and Charles Bergeron had a poster on trash detection in lakes, rivers and oceans. To conclude, the best student paper award went to a more theoretical paper by Kareem Amin, Michael Kearns and Umar Syed (U Penn) called “Bandits, Query Learning, and the Haystack Dimension” , which defines a measure of complexity for multi-armed bandit problems in which the number of actions can be infinite (there is some analogy to the role of VC-dimension in other learning models).
There were probably many other interesting posters worth being mentioned, but I didn’t have the chance to check them all!
On the personal side: my summer internship at NEC Labs with David Grangier is about to finish. It was an amazing learning experience and I am very grateful for it. Next step: back to Idiap Research Institute, EPFL and all the Swiss lakes and mountains! :)

Machine Learning recent sites

Tue, 06 Jul 2010 00:00:00 +0000

In the last few months (in which I haven’t posted in this blog) there were a few interesting web platforms related to Machine Learning poping-up, most notably:
MLcomp.org - you can upload your datasets and/or your algorithms, and experiments will run automatically. Then you can see statistics related to classifier performances and computation times. It is intended to help researchers and practitioners comparing different methods, and it works as a collaborative platform where code and data can be shared.
MetaOptimize.com - it contains a great QA about Machine Learning and related topics, using the same web platform StackOverflow has for programming topics.
I find these two websites a great way to improve collaboration among the ML community. Highly recommended!
The latest link is more market oriented, and it comes from Google:
Google Predict API: it puts together well established ML algorithms in an API that developers can use to make predictions on their own datasets.

Optimism as Artificial Intelligence Pioneers Reunite

Tue, 08 Dec 2009 00:00:00 +0000

Just a short link to an article of the New York Times about AI .

In 1978, Dr. McCarthy wrote, “human-level A.I. might require 1.7 Einsteins, 2 Maxwells, 5 Faradays and .3 Manhattan Projects.”

I think we probably have the genius scientists around, but not so sure about the 0.3 Manhattan Projects!

Update: You might also want to read latest Shane Legg’s predictions about human level artificial intelligence .

TEDx Geneva

Mon, 07 Dec 2009 00:00:00 +0000

Today I assisted to the first edition of TEDx Geneva. This was a locally-organized event following the same spirit of the original TED talks: “ideas worth spreading”.
I think the program was really good, because in this region there are some many incredible organizations. He could listen to people from CERN, EPFL, the United Nations, the Red Cross and some independent Swiss adventurers and entrepreneurs. We also had the opportunity to (re)watch some videos of the most popular TED talks recorded in the US. All the speakers spoke in English, which in my opinion degraded the level of their presentations, simply because it’s not their native language. Even if one is relatively fluent, it’s always harder to make jokes and be entertaining. The event was also a bit too long, covering the full day.
Still, I greatly appreciated the experience and recommend it to others!

Choosing my tools

Wed, 11 Nov 2009 00:00:00 +0000

I’m doing research in the fields of Machine Learning and Computer Vision, so each time we have an idea for a new algorithm, I have to write code, run experiments and compare results. I have realized that the experimental part is really the bottleneck, we have more ideas than we can test. For this reason, it’s critical to chose a good set of tools you can work with. This is a list of my current choices, but I am continuously looking for more efficient tools.

Open PhD and Postdoc positions

Sun, 01 Nov 2009 00:00:00 +0000

My supervisor is leading a new European project called MASH, which stands for “Massive Sets of Heuristics”. There are open positions here in Switzerland, as well as in France, Germany and Czech Republic.
The goal is to solve complex vision and goal planning problems in a collaborative way. It will be tested in 3D video games and also in a real robotic arm. Collaborators will submit pieces of code (heuristics) that can help the machine solving the problem at hand. In the background, machine learning algorithms will be running to choose the best heuristics.
If you are interested in: probabilities, applied statistics, information theory, signal processing, optimization, algorithms and C++ programming, you might consider applying!

Gmail Machine Learning

Wed, 14 Oct 2009 00:00:00 +0000

I just quickly tried the new Gmail Labs feature “Got the wrong Bob”? and it actually works quite nicely! I put some email addresses of family members, followed by the address of an old professor, who has the same first name of one of my cousins, and… Gmail found it! :) It suggested right way to change to the correct person, based on context!The other new feature, called “Don’t forget Bob”, is probably simpler, but quite useful as well. As I typed names of some close friends, I got more suggestions of friends I often email jointly with the previous ones. I wonder if the models to run this feature are very complicated. Probably they are not. I guess one just has to estimate the probability of each email address in our contacts to appear in the “To:” field, given the addresses we have already typed. To estimate these, you just have to use a frequentist approach and count how many times this happened in the past. With this in hands, “Got the wrong Bob?” will notice unlikely email addresses and “Don’t forget Bob” will suggest likely ones that are missing.

Schools kill creativity

Mon, 05 Oct 2009 00:00:00 +0000

My good friend Miguel called my attention to a TED talk that you might also find interesting:

Ken Robinson: Do schools kill creativity?

Ken Robinson argues that “schools kill creativity”, because kids are not given the chance to discover their interests and talents. Since very soon, students get a negative reward for making mistakes, which makes them too risk averse. He goes further, saying that the educational system is built to create university professors, leaving the majority of the students behing along the way. More space should be given to other forms of expressing intelligence, such as the arts or sports. I strongly recommend this video. Besides the interest of the subject, the presentation is actually quite funny, it somehow resembles a British-style stand-up comedy!

(My) ideal society

Sun, 02 Aug 2009 00:00:00 +0000

Thoughts on Life

This essay will be in beta version, longer than any Google product. Some assumptions

I am an individual.
I care about my life and happiness
There are other entities around. To different degrees, animals, persons, robots (future) also have feelings and personal ambitions. I interact with them. They will influence my level of well-being.

Happiness

The really wise people realized that happiness comes from the inside and that it should be our natural state. Reality in itself is neutral: nothing is good or bad! If you are feeling bad is just because you are having negative thoughts about what’s happening.

Increasing the scope

Wed, 22 Jul 2009 00:00:00 +0000

In the past it happened that I didn’t publish some potentially interesting thoughts in this blog, just because they didn’t exactly fit the “about intelligence” topic.
I’m fed up of this self-imposed censorship. In the future the scope will be broader.

Personal productivity, happiness and optimization algorithms

Wed, 22 Jul 2009 00:00:00 +0000

I spend lots of time wondering about the best ways to be both more productive and happy. Curiously, I’m coming to the conclusion that this is exactly what I should not do.

Being productive, like being happy, requires living the present moment, not thinking about it.

If you want to complete a task, the best strategy is just doing it! You might start by setting up a plan, a sequence of smaller actions that lead you to your goal, but once you have this, just do it. Spending too much energy re-planning and judging yourself along the way is just counter-productive.

Machine Learning to AI

Wed, 06 May 2009 00:00:00 +0000

John Langford wrote a very interesting post on the failures of Artificial Intelligence research and why Machine Learning has been a safer bet. Read it here .

Google CADIE vs Wolfram Alpha

Wed, 01 Apr 2009 00:00:00 +0000

Google already has a tradition of April fool’s jokes : this year they are introducing an Artificial Intelligence brain!

They describe the development process of their so called CADIE : Cognitive Autoheuristic Distributed-Intelligence Entity like this:

“For several years now a small research group has been working on some challenging problems in the areas of neural networking, natural language and autonomous problem-solving. Last fall this group achieved a significant breakthrough: a powerful new technique for solving reinforcement learning problems, resulting in the first functional global-scale neuro-evolutionary learning cluster.”

Machine Learning artwork

Sat, 28 Mar 2009 00:00:00 +0000

Today I tried out a great site to generate tag clouds, it is called wordle.net. I rendered some images just by copy-pasting the text from wikipedia about machine learning .

The results were pretty cool and I guess one could print awesome t-shirts with them. What do you say?

This one became officially my computer wallpaper:

ACM Paris Kanellakis Theory and Practice Award 2008

Wed, 18 Mar 2009 00:00:00 +0000

The 2008 ACM Paris Kanellakis Theory and Practice Award was awarded to Corinna Cortes and Vladimir Vapnik “for the development of Support Vector Machines, a highly effective algorithm for classification and related machine learning problems”.

It’s not the first time this award is given to Machine Learning people. In 2004 it was awarded to Yoav Freund and Robert Schapire “for the development of the theory and practice of boosting and its applications to machine learning.”

Computer Vision vs Computer Graphics

Wed, 28 Jan 2009 00:00:00 +0000

If I had to explain what computer vision is all about, in just one snapshot, I would show you this:

Computer Graphics algorithms go from the parameter space to the image space (rendering), computer vision algorithms do the opposite (inverse-rendering). Because of this, computer vision is basically a (very hard) problem of statistical inference.
The common approach nowadays is to build a classifier for each kind of object and then search over (part of) the parameter space explicitly, normally by scanning the image for all possible locations and scales. The remaining challenge is still huge: how can a classifier learn and generalize, from a finite set of examples, what are the fundamental characteristics of an object (shape, color) and what is irrelevant (changes in illumination, rotations, translations, occlusions, etc.).
This is what is keeping us busy! ;)

Vapnik's picture explained

Wed, 28 Jan 2009 00:00:00 +0000

This is an extremely geek picture! :) Let’s try to explain it:

First of all, as many of you know, the gentleman in the picture is Prof. Vladimir Vapnik . He is famous for his fundamental contributions to the field of Statistical Learning Theory, such as the Empirical Risk Minimization (ERM) principle, VC-dimension and Support Vector Machines.

Then we notice the sentence in the board: it resembles the famous “All your base are belong to us ”! This is a piece of geek culture that emerged after a “broken English” translation of a Japanese video game for Sega Mega Drive .

Stationary Features - Google Tech Talk

Thu, 08 Jan 2009 00:00:00 +0000

François Fleuret, my PhD advisor, recently gave a talk about object detection at Google (Zurich offices). You can now see it online:

https://www.youtube.com/watch?v=-w72_VwSj6A&t=3s

If you wonder where my research will try to extend the work done so far, just go to minute 45:30!

Machine Learning Summer School

Wed, 24 Sep 2008 00:00:00 +0000

Held in Ile de Ré (France), 1-15th September, this school counted with some famous names within the Machine Learning and Artificial Intelligence communities: Rich Sutton (co-author of the widely adopted book on Reinforcement Learning), Isabelle Guyon (co-author of the first paper on Support Vector Machines) and Yann LeCun (known for the convolutional neural network, energy based models and the DjVu image compression technique).

You can check the (almost) complete list of lecturers here . I found the course given by Shai Ben-David , on the Theoretical Foundations of Clustering" quite interesting and intriguing. Clustering seems to be really lacking solid theoretical support, which is surprising, given the importance of the problem. Some atempts are being done to axiomatize it, but there are a lot of open questions : what exactly is the class of clustering algorithms? how can you compare different clustering algorithms? why is a partition better than other?
Hope to see more developments in this area in the coming years.

ICVSS 2008

Tue, 22 Jul 2008 00:00:00 +0000

Last week I attended the International Computer Vision Summer School in Sicily, Italy. The main topics were Reconstruction and Recognition. I think the quality of the lectures, organization and location were all quite good, therefore I would recommend it to other PhD students.

Here is a short summary of some of the things we heard about:

Andrew Zisserman (Oxford, UK) - gave an overview of object recognition and image classification, with focus on methods that use “bag of visual words” models. Quite nice for newcomers like me!

Moved to Switzerland

Mon, 07 Jul 2008 00:00:00 +0000

Since the 1st of July, I am a PhD student at Idiap Research Institute and the Ecole Polytechnique Fédérale de Lausanne .
I am working in Machine Learning and Computer Vision under the supervision of Dr. François Fleuret .

Generating all possible pictures

Sat, 31 May 2008 00:00:00 +0000

Think of an image of 800 x 600 pixel and 24 bit of color (8 bit per each RGB component). Its trivial binary representation is a sequence of 11520000 bits (800 x 600 x 24) and we can think of each picture as being a natural number.

Imagine now that we write an computer program that generates all these pictures one by one, incrementing the natural number by one in each round.

Monkey with robotic arm

Thu, 29 May 2008 00:00:00 +0000

I’m not sure it’s recent news, because there is a public release from as back as 2005, but I just came across this video of a monkey eating using a robotic arm directly controlled by his brain. Researchers are from the Pittsburgh University.

Really impressive, although probably a bit tough for the monkey.

How Ant Colonies Get Things Done

Tue, 13 May 2008 00:00:00 +0000

Here you have a nice and very informative Google Tech Talk by Dr. Deborah Gordon on how ant colonies work without any central control:

It seems that ants make most of their decisions just based on the frequency they encounter other ants (which have a specific smell according to their role in the colony).

The amazing intelligence of crows

Tue, 13 May 2008 00:00:00 +0000

In this 10min TED talk, Joshua Klein talks about crows and how they are incredibly good learners.

They seem to have a powerful memory, use vision effectively, have problem solving skills, use tools and even learn from examples of other crows. I guess AGI is more than achieved at “crow-level Artificial Intelligence”!!

Science in Summer time

Fri, 09 May 2008 00:00:00 +0000

If everything goes as planned this year I am attending two Summer Schools.

The first one, the International Computer Vision Summer School 2008 , will be hosted in Sicily, Italy in 14-19 July. The program seems to be quite good and it will cover topics like object detection, tracking or 3D reconstruction, among others. There’s also a reading group on “how to conduct a literature review and discover the context of an idea”. The challenge is to see how far back in the past one can track the origins of a scientific idea. For example, the AdaBoost is a well known machine learning meta-algorithm, in which a sequence of classifiers is progressively trained focusing on the instances misclassified by previous classifiers. The set of classifiers is then combined by a weighted average. It was introduced by Freund and Schapire in 1996. This is easy to track, the question however is: can you find the same or similar core idea, or intution, somewhere else back in the past? Possibly from a different domain?
It’s gonna be fun!

How difficult is Vision?

Mon, 14 Apr 2008 00:00:00 +0000

Lately I have been wondering about the problem of Vision and how difficult it should be compared to problem of Artificial General Intelligence.
It seems to me that, given the order that it happened in Nature, processing visual input should be much simpler than using language or reasoning. I say this because there are quite simple animals with eyes, say a fish, a frog or a mouse… As I am not a biologist or neurologist, I am not sure what kind of visual tasks these animals are able to perform. For example, can a mouse tell if there is a cat in a picture or not?
In any case, I guess that these neuronal systems, much simpler than the human brain, are able to solve tasks that we have not yet achieved with Computer Vision algorithms.

Videolectures.net

Wed, 12 Mar 2008 00:00:00 +0000

A fast recommendation to the people interested in assisting to video talks in their computers:

http://www.videolectures.net

The website is specially interesting for the Machine Learning community, given that it has currently almost 600 videos on the topic, however you may find many other nice subjects. In fact if you work on Machine Learning, most likely I am not telling you anything new. In this case you could perhaps post a comment pointing to a video lecture that you found specially relevant. Thanks!

The First Conference on Artificial General Intelligence (AGI-08)

Fri, 22 Feb 2008 00:00:00 +0000

I will not be there, but I am looking forward to seeing what comes out of it.

The First Conference on Artificial General Intelligence (1st-3rd March, Memphis, US):

http://www.agi-08.org

Note that you can already read the submitted papers in the website.

Again on Measuring Machine Intelligence

Wed, 13 Feb 2008 00:00:00 +0000

I have recently found two tech reports written by Shane Legg and Marcus Hutter (IDSIA, Lugano, Switzerland) in which they make very interesting reviews on the definitions of machine intelligence and ways to measure it.
Have a look at:

Tests of Machine Intelligence
http://www.idsia.ch/idsiareport/IDSIA-11-07.pdf

Universal Intelligence: A Definition of Machine Intelligence
http://www.idsia.ch/idsiareport/IDSIA-10-07.pdf

It’s a pleasure to see that some people face the fundamental problems of AI from the front!

NARS: Non-axiomatic Reasoning System

Mon, 11 Feb 2008 00:00:00 +0000

Pei-Wang’s well-defined approach to Artificial General Intelligence takes as basic premises the fact that the agent has limited time and memory resources. He then develops a reasoning system that learns from experience and is able to deal with uncertainty and contradictory data.

http://nars.wang.googlepages.com

The project has become open-source, so you can even have a look at the code. There is also an free e-book on the webpage.

Brain Science Podcast

Sun, 03 Feb 2008 00:00:00 +0000

Another suggestion of a podcast, this time in the field of neuroscience. It features interviews and book reviews. Have a look at:

http://brainsciencpodcast.wordpress.com

John Searle: Beyond Dualism

Wed, 30 Jan 2008 00:00:00 +0000

It’s now time to make the first suggestion about philosophy of mind. And who else could it be if not the well-known american philosopher John Searle (the one from the Chinese-room argument against strong AI).
Check out this animated talk at IBM Almaden Institute on Cognitive Computing (2006).

On Intelligence by Jeff Hawkins

Wed, 30 Jan 2008 00:00:00 +0000

I would like to recommend this book by Jeff Hawkins, in which the author tries to create a theory about the neocortex.

He claims that the neocortex is basically a hierarchical memory system able to detect temporal and spatial patterns. Jeff Hawkins, and his company Numenta , are now trying to move forward and implementing this “neocortical algorithm” as software running on a computer. I enjoyed a lot reading it and I am trying now to read the technical papers. So far it looks like a good model, specially for computer vision systems, but it’s not yet clear to me how to solve problems from other cognitive areas such as language processing or planning. More posts on that for the coming weeks!

Measuring Intelligence

Mon, 28 Jan 2008 00:00:00 +0000

In order to develop artificial intelligence further, it would be important to have a formal and quantitative way to measure intelligence of an agent, being it a human or a machine.
The most famous test for artificial intelligence is the so-called Turing Test , in which “a human judge engages in a natural language conversation with one human and one machine, each of which try to appear human; if the judge cannot reliably tell which is which, then the machine is said to pass the test”. There is even a competition, the Loebner Prize which really evaluates different chatbots and choses the one who most resembles a human. However, this test is nowadays considered to be anthropomorphically biased, because an agent can be intelligent and still not be able to respond exactly like a human. Marcus Hutter as recently proposed a new way of measuring intelligence, based on the concepts of Kolmogorov Complexity and Minimum Description Length, in which compression = learning = intelligence. The Hutter Prize measures how much one can compress the first 100MB of wikipedia. The idea is that intelligence is the ability to detect patterns and make predictions, which in turn allows one to compress data a lot.
In my opinion this is not yet a totally satisfactory way of measuring general intelligence, for at least two reasons:

Artificial General Intelligence

Tue, 15 Jan 2008 00:00:00 +0000

Back in 1956, the founders of the new AI research field (John McCarthy, Marvin Minsky, Allen Newell and Hebert Simon) were deeply convinced that in a period of one generation we would have human-level intelligent computers.
However, after more than 50 years, we are still not able to solve some tasks that humans do without any apparent effort (such as distinguishing a dog from a cat or a horse in any kind of picture). Many frustrating results mark the history of AI: low quality of (early) machine translation systems, lack of robustness of speech recognition and computer vision systems, etc.
The so called “AI winter” is generally perceived to be finished by now, since many researchers have new hopes on building Artificial General Intelligence. Recent contributions from both neuroscience and theoretical computer science were decisive to create this optimism.
Here is a book edited by Ben Goertzel and Cassio Pennachin putting together several of the different renewed ideas.

Talking Robots

Tue, 15 Jan 2008 00:00:00 +0000

Talking Robots is a “podcast featuring interviews with high-profile professionals in Robotics and Artificial Intelligence for an inside view on the science, technology, and business of intelligent robotics”.

This podcast is produced at the Laboratory of Intelligent Systems, EPFL, Lausanne, Switzerland and it comes out every two weeks.
In future posts we will comment some of the episodes. Stay tunned!

Welcome to "About Intelligence"

Tue, 15 Jan 2008 00:00:00 +0000

Welcome to the blog where you can find ideas, comments and reviews about Artificial Intelligence, Robotics, Neuroscience, Consciousness and Philosophy of Mind.
Looking forward to having your feedback!