Barcelona Supercomputing Center (BSC) offers a competitive PhD grant to fight Alzheimer’s with AI and supercomputation. This grant is addressed to candidates who can start doctoral studies at the Universitat Politecnica de Catalunya (UPC) in September 2020, for example, by holding a MSc degree. Applicants must obtain the grant through the Doctoral INPhINIT Fellowships Programme (Incoming) from “la Caixa” Foundation. Eligible candidates must possess a strong academic record and satisfy, among other requirements, the mobility rule: not having their residence in neither Catalonia nor Spain for more than 12 months during the past three years. The fellowship consists of a yearly gross salary of €34,800, plus €3,564 for attending to conferences or purchasing personal equipment. The research will be developed in the Emerging Technologies for Artificial Intelligence Group under the supervision of Prof. Jordi Torres and Dr. Xavier Giro-i-Nieto, in collaboration with Dr Maite Garolera (Consorci Sanitari de Terrassa) and Prof. Petia Radeva (Universitat de Barcelona).
Full details on the requirements, conditions and application process are available here.
Increases in life expectancy in the last century have resulted in a large number of people living to old ages and will result in a double number of dementia cases by the middle of the century. The most common form of dementia is Alzheimer disease which contributes to 60–70% of cases. Research focused on identifying treatments to slow down the evolution of Alzheimer’s disease is a very active pursuit, but it has been only successful in terms of developing therapies that eases the symptoms without addressing the cause. Besides, people with dementia might have some barriers to access to the therapies, such as cost, availability and displacement to the care home or hospital, where the therapy takes place. We believe that Artificial Intelligence (AI) can contribute in innovative systems to give accessibility and offer new solutions to the patients needs.
Therapies such as reminiscence, that stimulate memories of the patient’s past, have well documented benefits on social, mental and emotional well-being, making them a very desirable practice, especially for older adults. Reminiscence therapy in particular involves the discussion of events and past experiences using tangible prompts such as pictures or music to evoke memories and stimulate conversation. With this aim, this project aims at exploring deep learning architectures to be used to develop an intuitive, easy to use, and robust dialogue system to automatize the reminiscence therapy for people affected by mild cognitive impairment or at early stages of Alzheimer’s disease.
In particular, the goal of this project is developing an AI conversational agent that simulates a reminiscence therapist by asking questions about the patient’s experiences. Questions will be generated from pictures provided by the patient, which contain significant moments or important people in user’s life. The activity pretends to be challenging for the patient, as the questions may require the user to exercise the memory.
A preliminary work by the research group is available on arXiv: Caros, Mariona, Maite Garolera, Petia Radeva, and Xavier Giro-i-Nieto. “Automatic Reminiscence Therapy for Dementia.” Master thesis at ETSETB TelecomBCN, UPC (2019).
- Develop an AI conversational agent implemented as a deep neural network capable of communicating with mild cognitive impairment through speech, and establishing a natural and engaging conversation based on the personal photographic records of the patient.
- Collect a dataset of therapist-patient conversations capable of training the AI agent as a reminiscence therapy, exploiting existing public dataset for pre-training the deep neural network for similar dialog tasks.
- Collaborate with clinical specialists to run experiments on patients following the best ethical practices, and propose an evaluation metric capable of quantitatively compare the performance of the AI agent with respect to a human therapist.
- Experimental evaluation of designed models using supercomputing systems, tuning model hyperparameters and configuration parameters to scale the training process.
- Publish the resulting work following the basic principles on open science, such us posting high quality papers on arXiv, software on GitHub and datasets in open repositories. The candidate will also be responsible to prepare a project page for its dissemination. High impact conference venues will be the primary target for publication: NIPS, ICLR, ICML, CVPR, or similar. Extensions of these works will be submitted to journals.
- Participate in the research team activities, such as group meetings and reading groups.
- Advanced programming in Python.
- Understanding of deep learning technology and motivation on turning deep learning technologies into practical.
- Experience in deep learning frameworks: PyTorch or TensorFlow…
- Comfortable in the use of UNIX-like operating systems.
Application process are available here.
(Deadline for submitting applications: 4 February 2020)