Presentations and Public Media given in recent years
Agents, Planning, and Theories of Mind given on 10 September 2024 at King's College London.
Abstract: This talk discussed my recent work on the intersection of agent reasoning and automated planning. This includes the integration of planning within an agent's practical reasoning mechanism, and how such integration provides a pragmatic theory of mind when combined with goal recognition algorithms. Throughout the talk, I will highlight advances and point towards opportunities for further research.
Goal Recognition through Reinforcement Learning
given on 13 May 2024 at the Reasoning and Learning for Autonomous Agents Symposium for the Agents Vic research group in Melbourne, Australia.
Abstract: Goal Recognition aims to infer an agent's goal from a sequence of observations. While most existing approaches often rely on manually engineered domains and discrete representations, recent work has tried to obviate the need for such domains using machine learning. This talk covers recent methods we developed towards model free reinforcement learning applied to the problem of goal recognition. We start with tabular methods that show the viability of the such techniques and then show how we can employ deep reinforcement learning methods to overcome the usual problems of partial observability and noise. We proceed to discussing the interplay of such methods with agent architectures pointing to the future of research on machine learning and reasoning.
Três mal-entendidos envolvendo inteligência artificial: an opinion piece I wrote for the largest newspaper in my home city of Porto Alegre, in Brazil. This piece explains the recent advances in AI trying to debunk myths about (supposed) dangers, while explaining where the dangers really lie, while demystifying the area cutting the marketing hype.
A vida - de Doutorando em Computação - como ela é (the life of a CS PhD student as it is), a panel I chaired with four PhD students in our PhD program, discussing motivations, standard of living, among other topics.
LP-Based Approaches for Goal Recognition as Planning, invited talk for the Agents Victoria group in Melbourne, Australia. It summarises my most recent push into goal recognition algorithms.
Abstract: Goal recognition is the task of inferring the intended goal of an agent given a sequence of observations. Advances in heuristics based on linear programming allows us to solve goal recognition tasks by encoding the declarative knowledge about such tasks resulting in two central contributions. First, we develop an approach that guarantees we select the actual hidden goal given the complete sequence of either optimal or suboptimal observations. Second, we automatically estimate the number of missing observations through a metric of uncertainty, which improves accuracy under very low observability. Experiments and evaluation show that the resulting approach is fast and dominates previous methods providing lower spread and higher accuracy on average.
Goal Recognition with Real-World Data, keynote for the Brazilian Conference on Intelligent Systems given in October 2019 at Salvador, Bahia.
It summarises my research from 2015 to 2019 on plan and goal recognition.
Abstract: Plan and goal recognition is the task of inferring the plan and goal of an agent through the observation of its actions and its environment and has a number of applications on computer-human interaction, assistive technologies and surveillance.
Recent work on such techniques using planning domain theories require almost perfect engineering of the domain theory and usually rely on assumptions of observability being either full, noise-free, or both.
These assumptions are too strong in the real world, regardless of the technique used to translate sensor data into symbolic logic-based observations.
In this talk, we discuss plan recognition techniques based on classical planning domain theories gradually relax all of these assumptions by leveraging advances on both symbolic planning and machine learning.
Inteligência Artificial não é (só) Aprendizado de Máquina, presented as part of the Dito Efeito series of Talks on Artificial Intelligence: science, business and challenges (in Portuguese). Presented in Porto Alegre on June 19th, 2019.
Plan and Goal Recognition in the Real World, given in October 2018 to undergraduate students at the Catholic University of Pelotas.
How to write a (good) research paper, a few tips on writing papers, adapted from the much better presentation from Simon Peyton-Jones (Microsoft Research), which I have given to an seminar series at PPGCC.
Plan and Goal Recognition in the Real World. This is an invited talk given at the University of Edinburgh, and the University of St. Andrews on September 2017. It summarises my research from 2015 to 2017 on plan and goal recognition.
Abstract: Plan and goal recognition is the task of inferring the plan and goal of an agent through the observation of its actions and its environment and has a number of applications on computer-human interaction, assistive technologies and surveillance.
Although such techniques using planning domain theories have developed a number of very accurate and effective techniques, they often rely on assumptions of full observability and noise-free observations.
These assumptions are not necessarily true in the real world, regardless of the technique used to translate sensor data into symbolic logic-based observations.
In this work, we develop plan recognition techniques, based on classical planning domain theories, that can cope with observations that are both incomplete and noisy and show how they can be applied to sensor data processed through deep learning techniques.
We evaluate such techniques on a kitchen video dataset, bridging the gap between symbolic goal recognition and real-world data.
How to write a (good) research paper, a few tips on writing papers, slightly adapted from the much better presentation from Simon Peyton-Jones (Microsoft Research), which I have given recently at PUCRS in the computer science academic week
Programming Autonomous Behaviour: Abstractions and Techniques, presented at the University of Brasília for Célia Ghedini's AI class on the 15th of August, 2014.
Practical Normative Reasoning: Models and Challenges, presented at the University of Aberdeen for Nir Oren's research group on the 14th of February, 2013.
An approach to generate MDPs using HTN representations, presented at Imperial College London for Alessio Lomuscio's research group on the 24th of June, 2011.
Norms Working Group QA, presented at the Seventh International Conference on Autonomous Agents and Multiagent Systems.
Extending agent languages for autonomy, a quick summary of my PhD thesis, presented at PUCRS.
AI and Negotiation, a panel hosted by Katia Sycara (CMU), myself and Michael Lewis (UPitt).