1 / 15

Emotion-based Agents: putting the puzzle together

Institute for Systems and Robotics. Emotion-based Agents: putting the puzzle together. Rodrigo Ventura http://www.isr.ist.utl.pt/~yoda email: yoda@isr.ist.utl.pt. Emotions versus Rationality René Descartes, Discourse on the Method , 1637.

nonnie
Download Presentation

Emotion-based Agents: putting the puzzle together

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Institute for Systems and Robotics Emotion-based Agents:putting the puzzle together Rodrigo Ventura http://www.isr.ist.utl.pt/~yoda email: yoda@isr.ist.utl.pt

  2. Emotions versus Rationality • René Descartes, • Discourse on the Method, 1637. • Disembodied mind: reason proper separated from body proper • Emotions and Rationality • Antonio Damásio, • Descartes’ Error, 1994. • Emotion mechanisms take an important role in reasoning processes

  3. Research goals: To understand emotion mechanisms, w.r.t. the design of autonomous agents: • Coping with complex and dynamic environments • Capable of determining and using relevance • Intuition: emotions provide a rough assessment of a situation, which can be refined by cognitive processes

  4. Approaching emotions from twodifferent perspectives • External manifestations  social interaction • Kismo [Breazeal] • Affective computing [Picard] • Believable agents [Reilly] • HCI [Pelachaud]

  5. Approaching emotions from twodifferent perspectives • Internal manifestations  behavioral consequences • “Future myopia” [Damásio] • Alarm system [Sloman] • Appraisal theory [Frijda] • Category of perceptions [Arzi-Gonczarowski]

  6. SENSORY CORTEX high road SENSORY THALAMUS AMYGDALA low road emotional stimulus emotional response • Lessons from neurobiology • [LeDoux] high and low roads to the amygdala

  7. Lessons from neurobiology • [Damasio] somatic marker hypothesis • the association of certain sensory images with body states • e.g.: experiencing a gut feeling when a certain response option comes into mind, however fleetingly • lesions in the emotional circuitry of the brain lead to “future myopia,”i.e., inability to preview long-term consequences of one’s own actions

  8. cognitive image ic(t) stimulus s(t) perceptual image ip(t) • The DARE model: Double representation of stimuli • cognitive image - oriented towards recognition • "what is it?" complex, slow • perceptual image - feature extraction • "what to do?" simple, fast

  9. Illustrative example: handwritten digit recognition • binary images, 32x32 pixels stimuli and cognitive images: ic = s perceptual images: ip

  10. (Ip, Ic) action Ic(t) matches future future future Movie-in-the-brain • puzzle piece #1: • “Movie-in-the-brain” and the inverted pendulum experiment • stored sequence of frames consisting of(Ic, Ip) pair and ensuing action • the Ic(t) extracted from the present stimulus is matched against the stored sequence

  11. The inverted pendulum experiment • Perceptual level: bang-bang tunning • Cognitive level: “movie-in-the-brain” Perceptual level Perceptual + cognitive levels

  12. (3) ic+(t) ic(t) ip(t) (2) s(t) memory (1) 1. perceptual metric 2. cognitive metric 3. minimization Sp(t) • puzzle piece #2: • Metric spaces and the handwritten digit recognition experiment • Assume that the spaces of cognitive and perceptual images are metric spaces • Indexing mechanism

  13. Theoretical results: • Under certain circumstances, there are garantees that the best cognitive match is found, using the indexing mechanism • Experimental results: • Significative efficiency gain, using the indexing mechanism

  14. puzzle piece #3: • Work in progress: finding relevant features • Example: Pavlov conditioning why the bell? • Example: dataset of 2000 handwritten digits, with 649 features each  what are the relevant features for a correct classification? • Dimensionality reduction methods: PCA, NMF, LSA, MDS, etc.

  15. Research perspectives • GOAL: To construct a formal/theoretic model of an emotion-based agent • TOOLS: • Relevance • Conditioning • Chunking

More Related