IPAL Research Themes
Theme 1: Explainable and Trustable AI
This theme covers techniques that lead to AI model/systems that would lead to better human trusts towards outputs from an AI system, in particular where the learnt models can be explained in a human-understandable manner, and/or can be certified correct.
In the last decade, artificial intelligence and machine learning techniques had spectacular success in varied application domains. The most famous one is the classification of images by deep neural networks, but many other techniques (Reinforcement Learning…) have heavily impacted other application domains (control theory…). These tools are largely seen as black-box techniques, which work extremely well on average. As black box systems, however, they are not designed to be understandable. This creates a major lack of trust in their behavior: why do they take the decision they do? Can we believe blindly in their choices? This is particularly important in safety-critical systems, where a mistake can threaten lives. Lately, flaws have been discovered in deep neural networks, such as their brittleness and over sensitivity to small perturbations, or the presence of hidden biases (against minority, female, etc.) in their decisions.
Kuldeep S. Meel (NUS), Jin Song Dong (NUS-SoC), Reza Shokri (NUS-SoC), Blaise Genest (CNRS), Joao Marques-Silva (ANITI), Nicholas Asher (CNRS), Leila Amgoud (CNRS), Jean Michel Loubes (Univ. Toulouse), Hwee Kuan LEE (ASTAR)
Theme 2: AI & HCI (Man-Machine Symbiosis)
This theme deals with new HCI paradigms that integrate AI techniques, or AI techniques that involve human-in-the-loop interactions.
AI aims to devise an artificial rival to human intelligence. HCI focuses on conceiving new interaction paradigms and improving applications as they approach widespread use. These different goals led to distinct priorities in the past. HCI has drawn participation and inspiration from short term projects with direct impact on mass market products. AI has mainly focused on future possibilities, with slow progress.
In recent years, AI and Human-Computer Interaction (HCI) have converged. AI techniques have improved and are currently in the toolset of more and more HCI researchers. Applications of machine learning are largely visible in the HCI literature. On the other side, AI technologies that are maturing rely on the contribution of the HCI community. Indeed, some AI researchers work on useful, usable and trustable systems, where they need methods and tools from the HCI community.
Greater interaction across the two fields is all but inevitable.
There are two research domains where AI and HCI are strongly converging : human augmentation (or more specifically Human Enhancement Technologies) and Intelligent User Interfaces. Human Enhancement Technologies (HET) refer to temporary or permanent attempts to overcome the current limitations of the human body by natural or artificial means. It can be applied to restore or improve human motor, sensory or cognitive performances. An intelligent user interface (IUI) is a user interface that relies, in some aspects, on AI. Generally, IUIs rely on the knowledge of the aim, context, and/or user being involved in the interactive task. It allows the interface to better fit the user’s needs and personalize or guide the interaction. The following IPAL projects are in the field of HET or IUI.
Christophe Jouffrais (CNRS), Joo Hwee Lim (A*STAR), Wei Tsang Ooi (NUS-SoC), Benoit Cottereau (CNRS), Terence SIM (NUS-SoC), Axel Carlier (ENSEEIHT), Vincent Charvillat (ENSEEIHT), Christophe Hurter (ENAC), Ying SUN (A*STAR), Jean-Pierre CHEVALLET (UGA), Philippe MULHEM (UGA), Georges QUENOT (UGA), Geraldine MORIN (ENSEEIHT), Brian Lim (NUS-SoC), Jamie Ng (A*STAR-I2R)
Theme 3: Natural Language Processing
The Natural Language Processing (NLP) is a multidisciplinary field involving linguistics, computer science and artificial intelligence, which aims to create natural language processing tools for various applications. In the last decades, the NLP research domain benefited from the strong improvements
made in AI tools and methods, and sometimes led to novel approaches based on neural networks.
Farah Benamara (Univ Toulouse), Nancy CHEN (A*STAR-I2R), Nicholas ASHER (CNRS), Luu Anh Tuan (A*STAR-I2R), Jian SU (A*STAR-I2R), Sahar Ghannay (CNRS), Pierre Zweigenbaum (CNRS), Anh Tuan LUU, Bin CHEN (A*STAR-I2R)
Theme 4: Data Science and Applications
The Data Science and Applications theme focuses on data analysis techniques, data management and their applications.
Bogdan Cautis (Université Paris-Sud), Caroline Chaux (CNRS), Cédric Févotte (CNRS), Vincent Y. F. Tan (NUS), M. KORDA (CNRS), J.B. LASSERRE (CNRS), V. MAGRON (CNRS), Kim TOH (NUS), Savitha Ramasamy (A*STAR), Martial Mermillod (CNRS), Xiaokui Xiao (NUS-SoC), Christophe Jouffrais (CNRS), Adrian Chong, (NUS), Benoit Delinchant (UGA), Savitha Ramasamy (A*STAR), Basura Fernando (A*STAR), Talel Abdessalem (Télécom Paris), Pierre Senellart (CNRS), Stéphane Bressan (NUS).
Theme 5: Efficient AI
This theme covers techniques that aim to improve the computational efficiency of AI, through improved hardware, mathematical and algorithmic techniques, or systems design. Examples include neuromorphic or bio-inspired techniques.
Wei Tsang OOI (NUS), Sylvie CHAMBON (ENSEEIHT)