Héctor Palacios

Hector Palacios - hectorpal     

Héctor Palacios

hector DOT palacios -AT- elementai -DOT- com
hectorpal -AT- g mail -DOT- com

About me

Hector is a Research Scientist at Element AI, a startup specialized in AI. He currently researches NeuroSymbolic methods, combining Machine Learning and Reasoning with two goals. One is about using business context/knowledge to enhance the performance of pure machine learning methods. The second is about incomplete but sound AI reasoning methods that deal with incomplete knowledge/models, including the output of pre-trained ML models. For reasoning, Hector is interested in constraints, SAT, planning, and causal models. His industrial work has focused mostly on NLP, although I have work with Vision, time-series and operations research. Back in academia, Hector produced award-winning work on AI reasoning, introduced new algorithms, and specialized in combining AI techniques.

More details in my Resume, or my publications. You could also be interested on my outdated CV (2016), or my Linkedin.

Before joining the industry, was a visiting professor at Universitat Pompeu Fabra, in the Artificial Ingelligence Group. In 2009, obtained a PhD on Informatics at Universitat Pompeu Fabra in Barcelona/Spain, under supervision of Hector Geffner. Most of the software related to publications or the International Planning Competition is available below. If you needed something not listed or find any bug, email me.

Research Interest

  • NeuroSymbolic methods for enforcing properties on the output of ML, and for solving simple AI reasoning problems where part of the model is from the output of a ML model.
  • Machine learning in combinatorial domains, including like natural language, reinforcement learning and supervised/unsupervised learning in rich domains.
  • High level Natural Language Task with structured output: question answering, dialogue, model adaptation
  • Automated planning under incomplete information: Conformant, contingent, probabilistic, non-deterministic planning.
  • Propositional modelling, Knowledge compilation languages: DNNF, OBDD
  • Tractability (theoretical and empirical) of crips models, including Planning, SAT and QBF


Selected Publications


Awards and Honors