SLIDE 1 Bild 1 Bild 1
FROM BIG DATA TO SMALL ROBOTS
CURRENT TRENDS OF AI AND OUR PLACE AS HUMAN USERS
ERIK BILLING – UNIVERSITY OF SKÖVDE
U N I V E R S I T Y O F S K Ö V D E – W W W . H I S . S E / E NFROM BIG DATA TO SMALL ROBOTS CURRENT TRENDS OF AI AND OUR PLACE - - PowerPoint PPT Presentation
FROM BIG DATA TO SMALL ROBOTS CURRENT TRENDS OF AI AND OUR PLACE - - PowerPoint PPT Presentation
FROM BIG DATA TO SMALL ROBOTS CURRENT TRENDS OF AI AND OUR PLACE AS HUMAN USERS ERIK BILLING UNIVERSITY OF SKVDE Bild 1 Bild 1 U N I V E R S I T Y O F S K V D E W W W . H I S . S E / E N Bild 2 Bild 2 Bild 3 Bild 3 Bild 4
SLIDE 2 Bild 2 Bild 2
SLIDE 3 Bild 3 Bild 3
SLIDE 4 Bild 4
SLIDE 5 Bild 5
19 September 2019 5
Goals and methodology 3/4
SLIDE 6 Bild 6
19 September 2019 6
In DREAM
“
We are taking this further
Robot assisted therapy
19 September 2019 6 SLIDE 7 Bild 7
19 September 2019 7
- 1. Is this a good method for
treating children with autism?
- 2. Technical aspects
1. Sense signals from the child 2. Detect what the child is doing 3. Make the robot react in a suitable way 4. Define subjective notions of “attention” and “imitation” so that the robot can understand?
Research questions
Erik Billing, www.his.se/erikb 7Cao et al. (2019) IEEE Robotics & Automation
SLIDE 8 Bild 8
19 September 2019 8
Ø AI is used to interpret and assess children's behaviour, and to control the robot Ø The system is designed with detailed input from clinicians as a tool for therapists Ø This is possibly
- nly close
collaboration between therapists and engineers
In sum
Erik Billing, www.his.se/erikb 8 SLIDE 9 Bild 12
Generation of new compounds that have attractive properties On molecular level
DEEP LEARNING FOR DRUG DESIGN
Erik Billing, www.his.se/erikbEfficacious Safe Minimal side effects
Polar surface area (PSA) molecular weight (MW) Lipophilicity (clogP)
Sthål et al. (2019) J. Chem. Inf. Model.
SLIDE 10 Bild 13
DEEP LEARNING FOR DRUG DESIGN
Erik Billing, www.his.se/erikbSthål et al. (2019) J. Chem. Inf. Model.
SLIDE 11 Bild 14
DEEP LEARNING FOR DRUG DESIGN
Erik Billing, www.his.se/erikbSthål et al. (2019) J. Chem. Inf. Model.
SLIDE 12 Bild 15
DEEP LEARNING FOR DRUG DESIGN
Erik Billing, www.his.se/erikb ) Molecular weight (b) clogP (c) PSA distribution of molecular properties in the original dataset s that are generated in the last 10 epochs (red). The target (a) Molecular weight (b) clogPSthål et al. (2019) J. Chem. Inf. Model.
SLIDE 13 Bild 16
Prediction of fungal infestation on oat
INFOFUSION FUSARIUM
Erik Billing, www.his.se/erikb SLIDE 14 Bild 17
17
SLIDE 15 Bild 18
We recomment that the “role of advisors and AgriDSS in advisory situations is reconsidered, changing from focusing on decision-making events/outputs towards thinking in terms of learning how to improve farmers situated seeing, and care”
FARMERS’ SITUATED KNOWLEDGE
Erik Billing, www.his.se/erikbLundström & Lindblom (2018) J. Agr. Sys.
SLIDE 16 Bild 23
Design by AI
- Decision support in design
- Drug design, Industrial settings, ergonomics
Open AI
- Data privacy
- Data lock in
Interaction with intelligent systes
- Transparent and Explainable AI
- User Experience Design
NEXT STEPS
Erik Billing, www.his.se/erikb SLIDE 17 Bild 24
19 September 2019 24 Erik Billing, www.his.se/erikb 24
Tack!
- https://www.his.se/en/sail/
- https://www.his.se/en/Research
/informatics/Interaction-Lab/