Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Illusory          Alastair Smith                  Supervised byConjunction and   Dr Eirini                  Mavritsakithe ...
RESEARCH AIMS / PLACEMENT OBJECTIVES Research Aims & Placement Objectives:   Adapt sSoTS model to incorporate Feature Bi...
COMPLEXIT Y OF VISUAL STIMULI Size                                                                            Shape       ...
A BRIEF EXAMPLE• Digits:         0–9• Letters:        TSNOX• Colours: PINK, YELLOW, GREEN, BLUE, BROWN• Treisman (1982)   ...
4   N   T   O   8                5
DIGIT   LETTER   LETTER   LETTER   DIGIT                                     6
7
4   N   T   O   8                8
9
BACKGROUND: ILLUSORY CONJUNCTION Treisman & Schmidt (1982)                                            4    N      T      ...
BACKGROUND:            FEATURE INTEGRATION THEORY Treisman & Gelade (1980):   Features processed in parallel, early and ...
SPIKING SEARCH OVER TIME AND SPACE MODEL                     (sSoTS) Mavritsaki et al (2006)   Biologically plausible, s...
sSoTS: ARCHITECTUREMavritsaki et al (2006)                         Inhibitory & non-specific pools:                       ...
sSoTS: NEURONAL-LEVEL MODEL Deco & Rolls (2005) Integrate & Fire Neurons Synaptic Currents:   NMDA, AMPA, GABA Freque...
sSoTS: MEAN FIELD APPROXIMATION Spiking-level neurons complex and computationally expensive Mean Field Approximation:  ...
BINDING Feature Binding:   Features processed in separate cortical areas   Features across differing dimensions are bou...
GROUPING Perceptual Grouping:                                                                      Feature Map   Duncan ...
GROUPING IMPLEMENTATIONActivation of pool within                               Feature Map feature map breaches   Grouping...
BINDING IMPLEMENTATION Activation of pool within feature map breaches Binding                                             ...
SIMULATIONS Simulation           Stimulus Onset           Stimulus Removed   0            100                 300 / 350 /...
BASELINE: SUCCESS RATE An item was identified in either position 3 or 4 in all trials, for  all displays and for both mod...
BASELINE: SUCCESS RATE                              Features conjoined in identified items Grouping & Binding: OFF       ...
BASELINE: REACTION TIME                                                  Mean Reaction Times                              ...
BASELINE: AVERAGE ACTIVATION Average activation across entire trial   Charts display average activation for „different‟ ...
BASELINE: SPIKE RATE Spike rate in first 1000ms   for „Same Shape‟ condition, trial 75,        H Map                    ...
BASELINE: SPIKE RATE Spike rate in first 1000ms   for „Same Shape‟ condition, trial 75,        Location Map             ...
BASELINE: SUMMARY Summar y:   G & B Model matches baseline success rates and RTs   G & B Model displays consistently qu...
IC CONDITIONS Stimuli presented for 250ms   0                                 100ms                     350ms            ...
IC CONDITIONS                             Mean Reaction Times            900.0            800.0            700.0          ...
IC CONDITIONS Grouping and Binding did not lead to a consistent increase in Illusor y  Conjunctions Increases in Illusor...
PRESENTATION PERIOD Is the stimulus being presented for too long?   This may explain why all models show a far greater p...
PRESENTATION PERIOD                                     Proportion of Illusory Conjunctions                               ...
COUPLING Is the self excitation (coupling) value set too high?   By reducing coupling the influence of external inputs t...
COUPLING                                   Proportion of Illusory Conjunctions                                        Prop...
LOCATION MAP FEEBACK Is the feedback from the location map too strong?   Binding increases feedback, therefore if feedba...
LOCATION MAP FEEBACK                                  Proportion of Illusory Conjunctions                                 ...
„BEST‟ MODEL Having run all possible combinations of the parameters  discussed, the following model produced the highest ...
 Contrast Map across first 1000ms („Dif ferent‟ condition, trial 50)               OFF                                ON ...
 Contrast Map across first 1000ms („Dif ferent‟ condition, trial 50)               OFF                                ON ...
FUTURE STUDY Low levels of Illusor y Conjunction error, may be due to method of analysis    Time locked analysis may pro...
Upcoming SlideShare
Loading in …5
×

Modelling Illusory Conjunction With S So Ts

724 views

Published on

  • Be the first to comment

  • Be the first to like this

Modelling Illusory Conjunction With S So Ts

  1. 1. Illusory Alastair Smith Supervised byConjunction and Dr Eirini Mavritsakithe sSoTS Model
  2. 2. RESEARCH AIMS / PLACEMENT OBJECTIVES Research Aims & Placement Objectives:  Adapt sSoTS model to incorporate Feature Binding and Grouping  Analyse the performance of the adapted model in the context of behavioural data from the Illusory Conjunction literature  Do results provide insight into the question:  How is grouping and binding implemented in the visual system and what are its effects? Background Background sSoTS Adaptation Study Results Discussion 2
  3. 3. COMPLEXIT Y OF VISUAL STIMULI Size Shape CircleColour Location Black Dimensions: Shape, Colour, Orientation Features: Circle, Red, 45° Background Background sSoTS Adaptation Study Results Discussion 3
  4. 4. A BRIEF EXAMPLE• Digits: 0–9• Letters: TSNOX• Colours: PINK, YELLOW, GREEN, BLUE, BROWN• Treisman (1982) Background Background sSoTS Adaptation Study Results Discussion 4
  5. 5. 4 N T O 8 5
  6. 6. DIGIT LETTER LETTER LETTER DIGIT 6
  7. 7. 7
  8. 8. 4 N T O 8 8
  9. 9. 9
  10. 10. BACKGROUND: ILLUSORY CONJUNCTION Treisman & Schmidt (1982) 4 N T O 8  Stimuli presented for 200 - 100ms (mean 120ms)  Instructed not to guess but to report “only what they were fairly confident they had seen” Mean number reported per trial Correct Item 0.52 Illusory Conjunction 0.39  Conjunction errors significantly exceeded errors which combined one correct feature with one not present in the display  Prinzmetal et al 1999; Gillbert & Humphreys, 2010  Conclusions:  When attention is loaded subjects make conjunction errors in dimensions that include colour, shape, size and solidity Background Background sSoTS Adaptation Study Results Discussion 10
  11. 11. BACKGROUND: FEATURE INTEGRATION THEORY Treisman & Gelade (1980):  Features processed in parallel, early and automatically  Behavioural & Physiological evidence  (Shepard ,1964; Garner ,1974; De Valois et al, 1975)  Attention required to combine features correctly to form objects  Focal attention applied serially to stimulus locations  Features present at the same location are combined to form objects  Visual Search data: Feature vs Conjunction search (Treisman et al, 1977)  Conjunction of features conjoined randomly unless:  Attended to  Top-down:  Contextual information  Past experience Background Background sSoTS Adaptation Study Results Discussion 11
  12. 12. SPIKING SEARCH OVER TIME AND SPACE MODEL (sSoTS) Mavritsaki et al (2006)  Biologically plausible, spiking-level neural network model of visual attention  Attentional selection operates through biased competition Mavritsaki et al (2011)  Links single-cell to psychological data for visual attention  Constrained by visual search data  Simulates:  Physiological studies of single-cell activity  Whole system behaviour in visual search  Break-down in performance following neural lesioning  Data from brain imaging Background sSoTS sSoTS Adaptation Study Results Discussion 12
  13. 13. sSoTS: ARCHITECTUREMavritsaki et al (2006) Inhibitory & non-specific pools: Ratio between inhibitory and Location Map: excitatory neurons 20-80 same for Activation relative to all layers, derived from populations saliency of stimuli in the brain. (Rolls & Deco, 2002) (Pulvinar (Shipp, 2004)) Noise:Simulating input fromother brain areasfollows Poissondistribution , λ =3Hz Stimulus InputProvided at 150Hz tocorresponding pools A H Feature Maps: Record presence offeatures in the visualfield (regions V2 - V4) H A Background sSoTS sSoTS Adaptation Study Results Discussion 13
  14. 14. sSoTS: NEURONAL-LEVEL MODEL Deco & Rolls (2005) Integrate & Fire Neurons Synaptic Currents:  NMDA, AMPA, GABA Frequency Adaptation mechanism  Based on Ca 2+ sensitive K + current  Inhibits activation as Ca 2+ concentration increases Spiking-level neurons complex and computationally expensive  Mean Field Approximation allows modelling of large numbers of integrate and fire neurons  Deco & Rolls (2005); Brunel & Wang (2001) Background sSoTS sSoTS Adaptation Study Results Discussion 14
  15. 15. sSoTS: MEAN FIELD APPROXIMATION Spiking-level neurons complex and computationally expensive Mean Field Approximation:  Deco & Rolls (2005); Brunel & Wang (2001)  Allows modelling of large numbers of integrate and fire neurons  Grouped into populations who share statistical similarity  Activity of group represented by distribution of neural spikes  Assumption = Network stays in stationary state Background sSoTS sSoTS Adaptation Study Results Discussion 15
  16. 16. BINDING Feature Binding:  Features processed in separate cortical areas  Features across differing dimensions are bound if they inhabit the same spatial location within the visual stimulus  Binding: Association between location and corresponding features  Tales et al (2002); Gillebert & Humphreys (2010)  High levels of noise may lead to incorrect binding of features to location  Reduced stimulus exposure leads to impoverished input signal  Errors in feature binding may produce greater incidents Background sSoTS Adaptation Adaptation Study Results Discussion 16
  17. 17. GROUPING Perceptual Grouping: Feature Map  Duncan (1995)  Elements with shared feature can be selected or rejected as a group 4Hz  => increased activation selected group  Increased activation in grouped items 4Hz may lead to binding errors, which in turn may result in increased Illusory Conjunction error Background sSoTS Adaptation Adaptation Study Results Discussion 17
  18. 18. GROUPING IMPLEMENTATIONActivation of pool within Feature Map feature map breaches Grouping threshold 4Hz Is there another poolwithin the same feature No map that has breached 4Hzthe Grouping threshold? Yes Increase excitatory connections in both directions between Grouping Threshold = 4Hz flagged pools Background sSoTS Adaptation Adaptation Study Results Discussion 18
  19. 19. BINDING IMPLEMENTATION Activation of pool within feature map breaches Binding Binding Threshold = 4Hz threshold Is the same position in the location map already bound Yes NoDoes another pool exist associated with the same positionyet in a different feature dimension that has also breached No the binding threshold Yes Increase feedback location map to corresponding pools Background sSoTS Adaptation Adaptation Study Results Discussion 19
  20. 20. SIMULATIONS Simulation Stimulus Onset Stimulus Removed 0 100 300 / 350 / 400 2000 (Baseline 2000) time (ms) Stimuli Different (dif) Same Colour (sc) Same Shape (ss) H A H A H H 100 trials per condition Background sSoTS Adaptation Study Results Discussion 20
  21. 21. BASELINE: SUCCESS RATE An item was identified in either position 3 or 4 in all trials, for all displays and for both models  An item is said to be identified when the contrast between its activation and that of its distractors reaches a 60% threshold  Attention Index: Luce (1959, 1977) Different (dif) Same Colour (sc) Same Shape (ss) H A H A H H Background sSoTS Adaptation Study Results Discussion 21
  22. 22. BASELINE: SUCCESS RATE Features conjoined in identified items Grouping & Binding: OFF  Grouping & Binding: ON 100 93 100 96 90 90 80 80 70 70 59 58 60 50 60 50 46 50 % % 40 40 30 30 20 20 10 1 10 1 1 0 0 0 0 0 Dif SC SS Dif SC SS Condition Condition Correct Features Conjoined Illusory Conjunction Correct Features Conjoined Illusory Conjunction Grouping & Binding lead to an increas e in correctly identified items acros s all conditions and a s ing le increas e in Illus or y Conjunction error s in the s ame colour condition. Background sSoTS Adaptation Study Results Discussion 22
  23. 23. BASELINE: REACTION TIME Mean Reaction Times 470 465 460 Reaction Time (ms) 455 450 445 440 435 430 425 420 415 Dif SC SS Condition Off On Gro uping & Binding Mo del is quic ker to ident ify items Bo t h mo dels ident ify items slower when t here is no similarit y bet ween o bjec t s displayed Background sSoTS Adaptation Study Results Discussion 23
  24. 24. BASELINE: AVERAGE ACTIVATION Average activation across entire trial  Charts display average activation for „different‟ condition, blue map H A OFF  ON Peak = 2.5Hz Peak = 5Hz 1.5Hz 1.5HzCharacteristics co mmon to all maps :  Increase in peak activation  Activation similar until approx. 400ms  Exaggerated function:  Activation returns to similar levels by  steeper rate of increase and decline in approx. 950ms activation Background sSoTS Adaptation Study Results Discussion 24
  25. 25. BASELINE: SPIKE RATE Spike rate in first 1000ms  for „Same Shape‟ condition, trial 75, H Map H H OFF  ON Thresholds Breached 275-300ms 0 - 300ms: Both models show similar activity 275m: Pool 3 bound to Green Map 300ms: Grouping initiated between pool 3 and 4 450ms: Pool 4 bound to Blue Map Background sSoTS Adaptation Study Results Discussion 25
  26. 26. BASELINE: SPIKE RATE Spike rate in first 1000ms  for „Same Shape‟ condition, trial 75, Location Map H H OFF  ON Binding active for position 4 BH 450ms Binding active for position 3 GH Grouping active in H map 300ms 0-400ms: Models show similar activity 500–600ms: Effects of grouping in H map between position 3 and 4 present 650-850ms: Binding causes increased activation in position 4 Effect also present in trial in which grouping is not active Binding effect effects accumulative, therefore less immediate Background sSoTS Adaptation Study Results Discussion 26
  27. 27. BASELINE: SUMMARY Summar y:  G & B Model matches baseline success rates and RTs  G & B Model displays consistently quicker RTs for all displays  Activity in all maps accelerates at a greater rate in the G & B model during periods where binding is likely to be active  Binding is initiated before any acceleration in activity rate is displayed  Grouping and Binding leads to increased activity within bound and grouped pools , which corresponds to the period of accelerated activity in average activation plots  => Faster RTs result from an acceleration in activation caused by G & B  G & B model identifies a greater number of items correctly in all conditions and leads to an additional illusory conjunction error in the SC condition  When the period of exposure is reduced we expect the increased noise present in the system to lead to a greater proportion of incorrectly bound features, resulting in an increase in illusory conjunctions Background sSoTS Adaptation Study Results Discussion 27
  28. 28. IC CONDITIONS Stimuli presented for 250ms 0 100ms 350ms 2000ms Stimulus Onset Stimulus Removed 3% 7% % Illusory Conjunctions 6% % Illusory Features 5% 2% 4% 3% 1% 2% 1% 0% 0% Off On Off On Grouping & Binding Grouping & Binding  Similar proportion of Illusory features produced by both models  Small increase in proportion of illusory conjunctions produced by G&B model Background sSoTS Adaptation Study Results Discussion 28
  29. 29. IC CONDITIONS Mean Reaction Times 900.0 800.0 700.0 600.0 500.0 400.0 300.0 200.0 100.0 0.0 Dif SC SS Off On G & B model is consistently slower across all conditions Models display the same pattern of reaction times across conditions  Slowest Response = Different  Fastest Response = Same Shape Background sSoTS Adaptation Study Results Discussion 29
  30. 30. IC CONDITIONS Grouping and Binding did not lead to a consistent increase in Illusor y Conjunctions Increases in Illusor y Conjunctions were not a levels consistent with experimental evidence Number of Illusory conjunctions produced in each display condition (60% contrast threshold) Display Dif SC SS On 2 0 2 G&B Off 0 0 2 What factor s in the G&B model lead to additional Illusor y Conjunctions Background sSoTS Adaptation Study Results Discussion 30
  31. 31. PRESENTATION PERIOD Is the stimulus being presented for too long?  This may explain why all models show a far greater proportion of correct responses Is the stimulus being presented for too short a period?  Values for both illusory and correct responses are very low, by increasing the presentation period clearer trends may emerge Presentation periods:  200ms  further reductions lead to difficulty breaching 60% contrast threshold  250ms  Period used in initial shortened exposure trials  300ms  further increases lead to levels of performance similar to full 2000ms presentation Background sSoTS Adaptation Study Results Discussion 31
  32. 32. PRESENTATION PERIOD Proportion of Illusory Conjunctions Proportion of illusory features 3% identified % of Illusory Conjunctions 20% % of Illusory Features 2% 16% 12% 1% 8% 4% 0% 0% 200 250 300 200 250 300 Stimulus presentaion period (ms) Stimulus presentaion period (ms) Off On Off On P ro po r t io n o f illuso r y c o njunc t io ns low  T he pro po r t io n o f illuso r y feat ure s & fo r bo t h mo dels c o njunc t io n s inc reas e s as t he present at io n perio d dec reas ed G & B c o nsistent ly pro duc es marg inal ly g reater pro po r t io n o f IC s in all  Similar pro po r t io n o f illuso r y feat ures c o ndit io ns repo r ted by bo t h mo dels Background sSoTS Adaptation Study Results Discussion 32
  33. 33. COUPLING Is the self excitation (coupling) value set too high?  By reducing coupling the influence of external inputs to the pool is increased and fluctuations in these inputs has a more immediate effect  Therefore bound or grouped pools have a greater and more immediate impact on one another Values tested: Feature Map  Low: wplus = 2.2  lower values resulted a negligible number of items identified  Moderate: wplus = 2.3  High: wplus = 2.4  value used in all previous simulations Background sSoTS Adaptation Study Results Discussion 33
  34. 34. COUPLING Proportion of Illusory Conjunctions Proportion of Illusory Features 4% 12% % of Illusory Conjunctions % Illusory Features 3% 8% 2% 4% 1% 0% 0% Low Moderate High Low Moderate High Coupling (wPlus) Coupling (wPlus) Off On Off On P ro po r t io n o f IC s low ac ro ss all  P ro po r t io n o f illuso r y erro r s g reatest c o ndit io ns fo r bo t h mo dels when mo derate c o upling implemen t ed G & B mo del pro duc es g reater pro po r t io n o f illuso r y c o njunc t io n erro r s in all c o ndit io ns Background sSoTS Adaptation Study Results Discussion 34
  35. 35. LOCATION MAP FEEBACK Is the feedback from the location map too strong?  Binding increases feedback, therefore if feedback is already high pools that are bound will gain little advantage Values tested:  Low: wi4 = 0.15  lower values resulted a negligible number of items identified  Moderate: wi4 = 0.20  High: wi4 = 0.25  value used in all previous simulations Background sSoTS Adaptation Study Results Discussion 35
  36. 36. LOCATION MAP FEEBACK Proportion of Illusory Conjunctions Proportion of Illusory Features Identified 4% 16%% of Illusory Conjunctions % of Illusory Features 3% 12% 2% 8% 1% 4% 0% 0% Low Moderate High Low Moderate High Feedback (wi4) Feedback (wi4) Off On Off On G& B mo dels repo r ted mo re illuso r y  Hig her pro po r t io ns o f illuso r y feat ures c o njunc t io n s and illuso r y feat ures and illuso r y c o njunc t io n s are ident ified in t he mo derate feedbac k T he pro po r t io n o f illuso r y feat ure s and c o ndit io n c o njunc t io n s remain s low ac ro ss all c o ndit io ns  Bo t h mo dels display a similar t rend when feedbac k is adjusted Background sSoTS Adaptation Study Results Discussion 36
  37. 37. „BEST‟ MODEL Having run all possible combinations of the parameters discussed, the following model produced the highest propor tion of ICs: 12% 10%  Presentation Period: 250ms 8% 6%  Coupling: Moderate 4% 2% 0%  Feedback: High Illusory Illusory Features Conjunctions Off On Propor tion of Illusor y Conjunctions remains at a similar level Model per formance does not match behavioural data Background sSoTS Adaptation Study Results Discussion 37
  38. 38.  Contrast Map across first 1000ms („Dif ferent‟ condition, trial 50) OFF ON H AA MapH MapB l u e MapLocation Map Blue A Identified in Position 4 Background sSoTS Adaptation Study Results Discussion 38
  39. 39.  Contrast Map across first 1000ms („Dif ferent‟ condition, trial 50) OFF ON H AA MapH Map Time Lock AppliedB l u e MapLocation Map Background sSoTS Adaptation Study Results Discussion 39
  40. 40. FUTURE STUDY Low levels of Illusor y Conjunction error, may be due to method of analysis  Time locked analysis may prove a more appropriate measure of model performance Repor ted perception of features not present in the stimulus is commonly used as a baseline for analysis of illusor y conjunction error  Informs judgement of significance of 2 -3% illusory conjunction error produced by G&B model Feature – location association strongly coded within system from an early stage  Leads to few positions not receiving direct input breaching G & B thresholds  By increasing feedback from non -specific feature pools input signal can be blurred across locations  In conditions of high noise, number of cases in which incorrect features are bound should increase, leading to a corresponding increase in illusory conjunctions Background sSoTS Adaptation Study Results Discussion 40

×