Sign in

Ideas on artificial intelligence, and other goings-on, by

After posting last week on some notes regarding a workshop on deep learning meets logic, I came across this excellent article by Subbrao Kambhampathi. In broad strokes, he makes the following point (my interpretation): much of the recent emphasis in machine learning research has been on (big) data driven learning, with little regard to explicit knowledge, say in the form of symbols or constraints, and even treating the endeavour with scorn. He argues convincingly that this is ultimately a futile experience because there are many cases where leveraging available knowledge is effective.

There are four points that I think could…

Last week, Efi, Phokion, Loizos and I ran a virtual workshop (1) on deep learning & logic. The workshop was initiated many months ago by Efi; Sponsored by Samsung Cambridge, it finally came to life this month.

We learnt that over a thousand people registered for the workshop and had as many as 600 people attend on a single day. This was obviously very rewarding and gratifying to see.

I’ll gloss over the details of the full extent of the very exciting talks we heard. These details can be retrieved from the abstracts; besides the slides and the recorded videos…

The increasing amount of data, and the increasing complexity of the application context that is generating that data presents deep challenges to a machine learning practitioner.

On the one hand, we want to tease apart the model complexity using divide and conquer. Like programming language paradigms in the 70s and 80s, we’d like small chunks of code and models to work well with other chunks to realise the application pipeline.

Now, while languages like Python have access to an enormous library of machine learning modules, it requires significant programming experience to understand how the application should be broken down to…

When we think of commonsensical AI systems, we may expect them to infer obvious truths. For example, if we tell the system that penguins, although birds, don’t fly, and Aptenodytes forsteri (Emperor penguins in English) are a breed of penguins, the system should be able to figure out that these things don’t fly as well.

But logical truths may not be the result of such simple rule applications. Solving a sudoku puzzle, or even checking that something is a solution to a sudoku puzzle, without knowing that the logical sentences you are looking at is the logical encoding of that…

A few weeks ago, I attended a seminar on epistemic planning:

the seminar group
the seminar group
the seminar group

Automated planning in AI is concerned with computing a sequence of actions that achieves a goal. Epistemic planning is a flavour of planning where we explicitly represent and reason about the mental states of the other participants in the environment.

Part of the reason this paradigm is important is that in social environments, it can be frustrating to get canned responses where little attention is paid to the individual’s needs. Think of lengthy interactions between a robot and a person in hospital care. …

In a recent article, I prefaced the renewed emphasis on explainability in AI as follows:

In a seminal paper, McCarthy [1958] put forward a profound idea to realize artificial intelligence (AI) systems: he posited that what the system needs to know could be represented in a formal language, and a general-purpose algorithm would then conclude the necessary actions needed to solve the problem at hand. The main advantage is that the representation can be scrutinized and understood by external observers, and the system’s behavior could be improved by making statements to it.

The past 60 years have yielded numerous methods…

Probabilistic models are widely used in artificial intelligence and machine learning. When using these models, we begin by stipulating a distribution on a set of random variables, the latter standing for the observable or latent atoms of our application domain. However, more often than not, we find ourselves in the position of having to deal with the inescapable fact that our knowledge of the world is incomplete! For example, we cannot know how many individuals we will encounter in our lifetime, and we certainly cannot be expected to know the identity and traits of these individuals as and when we…

For the last two decades, there has been considerable work on defining probabilistic graphical models (PGMs) using syntactic devices from first-order logic. Markov logic networks, Bayesian logic programs, and many others marry finite-domain first-order logic with probabilities, which then is equivalent to a complex and involved PGM. PGMs have supercharged the application of statistical methods in language understanding, computer vision, medical diagnosis, and automated planning.

At first glance, this logical machinery seems only useful for convenient codification. Essentially, we are still embedded in a standard PGM setting, and the logical stuff is purely syntactic sugar.

I’ve always been more interested…

Vaishak Belle

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store