1. How does the brain decide what to look at next? John Findlay University of Durham (acknowledgements to Val Brown)

2. Saccades are quintessentially voluntary movements. The gaze selects informative detail but eye scans also appear random and arbitrary from Yarbus (1967)

3. Pathways involved in saccade generation

5. Visual pathways form a massively interconnected neural network

6. Retinotopic mapping is maintained through to the saccadic generator in the SC

7. Multiple interconnected maps offer the possibility of selection by biased competition “ Some kind of short-term description of the information currently needed must be used to control competitive bias in the visual system, such that inputs matching that description are favoured in the visual cortex.” (Desimone and Duncan, Annual Review of Neuroscience, 1995 )

8. Biased competition creates salience maps A salience map is a two-dimensional map in which a single scalar quantity (salience) is represented at each point. Biased competition results in similarity to the search target being represented as salience. Instantiated in various models Itti L and Koch C (2000). A saliency-based search mechanism for overt and covert shifts of attention. Vision Research, 40, 1489-1506. Hamker FH (2004). A dynamic model of how feature cues guide spatial attention. Vision Research, 44, 501-521.

9. Saccade target selection in the superior colliculus The SC is the main final site for selection of saccade destinations

10. Activity in the superior colliculus related to saccades Wurtz R H (1996). Vision for the control of movement. The Friedenwald Lecture Investigative Ophthalmology and Visual Science, 37, 2131-2145.

11. How does the brain decide what to look at next? During visual search, biased competition creates a salience map and processes, probably in the SC, select the point of highest salience to convert to an orienting saccade. Supported by detailed studies of saccades during visual search (Findlay, Vision Research, 1997; Motter and Belky, Vision Research, 1998a,b)

12. Task: Search for a red cross (Look at it) Findlay J M (1997). Saccade target selection during visual search. Vision Research, 37, 617-631

13. Properties of first saccades (Findlay, 1997) Short latency (~ 250 ms) – very similar for saccades to target and to distractor

14. Properties of first saccades (Findlay, 1997) Short latency (~ 250 ms) – very similar for saccades to target and to distractor Frequently (75%) on target when target is in inner ring, occasionally (26%) when target in outer ring..

15. Properties of first saccades (Findlay, 1997) Short latency (~ 250 ms) – very similar for saccades to target and to distractor Frequently (75%) on target when target is in inner ring, occasionally (26%) when target in outer ring. Incorrect saccades go preferentially to distractor sharing a feature with the target.

16. Monkey visual search (Motter & Belky, Vision Research, 38, 1007-1022; 1885-1815, 1998) Monkeys trained to search for a conjunction target (colour and orientation)

17. Saccade selection in visual search The conclusion in both the Findlay and the Motter & Belky studies was that the biased competition/ salience map approach provided the most satisfactory account of saccadic selection. In particular, no evidence for a rapid covert attentional scan (favoured by many psychologists). This conclusion was reached earlier in physiological studies of single cell responses in the visual system of primates carrying out search tasks FEFSchall & Hanes (1993) ITChelazzi, Miller, Duncan & Desimone (1993)

18. How does the brain decide what to look at next during visual search ? During visual search, biased competition creates a salience map and processes, possibly in the SC, select the point of highest salience to convert to an orienting saccade.

19. Selection from a salience map is basic Supplementary processes 1. Inhibition of return 2. Saccade pipelining 3. Strategies How does the brain decide where to look next? 4. Implicit learning 5. Contingent learning 6. Neuro-economics (Glimcher) 7. Task specific requirements for information acquisition (Land, Hayhoe, Ballard)

20. Selection from a salience map is basic Supplementary processes 1. Inhibition of return 2. Saccade pipelining 3. Strategies How does the brain decide where to look next? NO OTHER ATTENTIONAL SELECTION 4. Implicit learning 5. Contingent learning 6. Neuro-economics (Glimcher) 7. Task specific requirements for information acquisition (Land, Hayhoe, Ballard)

21. Selection from a salience map is basic How does the system avoid ‘salience loops’? (B is the most salient location when A is fixated, then A becomes the most salient when B is fixated) Salience map alone would give A > B > A > B..... Proposed answer - Inhibition of Return (IOR) An attended location is subject to some form of inhibition when attention is shifted elsewhere Klein R M and MacInnes W J (1999). Inhibition of return is a foraging facilitator in visual search. Psychological Science, 10, 346-352

22. The rings task Scan order partly specified Centre - red – free scan through blacks - blue Count target letters and make Yes/No response

23. The Rings Task 3 6 912 Scan through the rings, starting with the red and ending with the blue (Scans from 6 individuals)

24. Rings task - typical eye scan

25. Deviations from sequential scan BACKTRACK 1 BACKTRACK 2 BACKTRACK 1 OMISSION

26. Trials with deviations from sequential scan OmissionsBacktrack 1Backtrack 2Backtrack > 2 BR15 1 JP91569 LS61234 LW52954 PB539510 SL3137 Proportion 0.090.260.060.11 Error rate* 47 % (target omissions) 6 %15 %23 % * Error rate on trials with standard scan 4 %

27. Backtracking in visual search BACKTRACK 1 Found by other workers (Motter & Belky, 1998) (Peterson et al. 2001) IOR time course may relate to visual processing slower with increased processing demands, so not always immediate Backtracking sequences may be pre-planned (pipelined saccades) In A 1, B, A 2 fixation sequences, the B fixation was normal duration (228 ms) but A 1 and A 2 were both shorter than normal (~ 170 ms). The saccade following a backtracking sequence tended to follow the direction of the last saccade in the sequence.

28. Backtracking in visual search BACKTRACK 1 Found by other workers (Motter & Belky, 1998) (Peterson et al. 2001) IOR time course may relate to visual processing Backtracking sequences may be pre-planned (pipelined saccades) In A 1, B, A 2 fixation sequences, the B fixation is normal duration but A 1 and A 2 are both shorter than normal The saccade following a backtracking sequence tends to follow the direction of the last saccade in the sequence X

29. Visually-guided and memory- guided saccades How does the brain decide which? Hikosaka et al (2000) argue for basal ganglia pathway (blue route) Inhibitory effects on SC, others excitatory Separate sets of cells in caudate and in SNr are active during visually guided and memory guided movements.

30. Directional strategies COUNT THE DOTS (Convex Hull)

31. Directional strategies COUNT THE DOTS Directional strategies are one form of memory (Gilchrist & Harvey)

32. NO OTHER ATTENTIONAL SELECTION ‘Visual attention selects the saccade target’ Statement supported by the finding that visual information at the destination point of a forthcoming saccade receives preferential pre-processing (Deubel and Schneider, 1996; Kowler et al. 1995) Biased competition is a form of attentional selection but does not operate in a localised region of the visual field. Visual attention is commonly thought of as selection of a localised region.

33. ‘Visual attention selects the saccade target’ Why I don’t like this statement 1.It’s getting close to a homunculus view 2.Localised visual attention should be able to eliminate distractor interference.

34. Visual attention should be able to eliminate distractor interference. next saccade ?

35. Visual attention should be able to eliminate distractor interference. Attentional spotlight selects next target

36. Visual attention should be able to eliminate distractor interference. Attentional spotlight selects next target

37. Visual attention should be able to eliminate distractor interference. Attentional spotlight selects next target No effective spotlight

38. Visual attention should be able to eliminate distractor interference. Attentional spotlight selects next target No effective small spotlight Global effect

39. The Global Effect Saccades to neighbouring target pairs tend to land towards a centre-of-gravity position. Findlay, 1981, 1982; Deubel, 1982; Ottes, Van Gisbergen and Eggermont, 1984 Reliably found with onset stimuli: does it occur in free scanning?

40. How accurate are scanning saccades ? SACCADE TO CENTRE OF GRAVITY ?

41. Is there a global effect in free scanning ? Are saccades less accurate when there is a distractor present in the critical sector (as defined by Walker et al. 1997)? SubjectNo distractor With distractor JP16.943.3 LW14.936.7 PB20.438.0 SL22.855.1 Percentage of inaccurate saccades

42. Probability of inaccurate saccade for distractors in different locations relative to the target

43. Is accuracy higher follower longer fixations ? accuracy coding 4 32 1 Accuracy is highest following short duration fixations (although distractors still decrease it). This is the opposite to a speed-accuracy trade off.

44. Attentional selection and saccades Saccades during a free scan of a set of identical elements show the global effect. Thus no evidence here for a spotlight-like attentional selection. In most practical situations, elements are not identical; hence biased competition will act to reduce the global effect

45. How does the brain decide where to look next? Selection from a salience map is basic Supplementary processes influencing salience 1. Inhibition of return 2. Saccade pipelining 3. Strategies 4. Implicit learning 5.Etc. etc. etc.

46. THE END Thank you for your attention

48. Kowler et al (1995)

49. Do distractors that have been already scanned reduce accuracy ? accuracy coding 4 32 1 Accuracy is reduced both by scanned distractors and by new ones.

50. REPLICA TRIALS NON REPLICA TRIALS How replicable are scanning patterns ? (repeat run with one subject - different trial order) NEAR REPLICA TRIALS REPLICAS 26% overall, 68% ring count 3 REPLICAS and NEAR REPLICAS 45% overall

51. Replicability of directional selection

52. Saccade direction histograms Heuristic scannersStrategic scanners

53. Saccade direction change histograms

54. Saccade Landing Points Accuracy is largely independent of saccade size SACCADES WITH NO DISTRACTOR IN SECTOR Subject PB

55. Saccade undershoot 1% 5% 10%

56. Saccade variability : on-axis 10% 5% 1% 10% 5% 2%

57. Saccade variability : off-axis 5% 2% 1%

58. Corrective saccades in the multi-element scanning task

59. Oculomotor capture in the multi-element scanning task