Unifying Negative Priming Theories Hecke Schrobsdorff,2, Matthias Ihrke,3, Jörg Behrendt,3, Marcus Hasselhorn,3, Jan Michael Herrmann,2 hecke@nld.ds.mpg.de Bernstein Center for Computational Neuroscience Göttingen 2 University of Göttingen, Institute for Nonlinear Dynamics 3 University of Göttingen, Georg-Elias-Müller Institute for Educational Psychology BCCN Symposium 25.9.27
Outline Negative Priming 2 Negative Priming Theories 3 The General Model 4 Implementation of Theories
Aging Effects in Selective Attention Behavioral Experiments Theoretical Psychology EEG Recordings ERP Analysis Computational Modeling Advanced Averaging Single Trial ERPs Biology of Aging
Aging Effects in Selective Attention Behavioral Experiments Theoretical Psychology EEG Recordings See Poster #64 by Matthias Ihrke ERP analysis Computational Modeling Advanced Averaging Single Trial ERPs Biology of Aging
Aging Effects in Selective Attention Behavioral Experiments Theoretical Psychology EEG Recordings ERP Analysis Computational Modeling Advanced Averaging Single Trial ERPs Biology of Aging
Negative Priming Introduction Main Question: How do we filter out irrelevant information about our environment? Negative Priming (NP) is a way to acces mechanisms of selective attention. is a slowdown of reaction time when responding to previously irrelevant stimuli.
Negative Priming Different Priming Conditions CO NP NP2 time CO PP PP2 CO stimulus onset reaction response stimulus interval reaction time time RTs vary with the order of subsequent stimuli. NP2 > NP > CO > PP > PP2 in general.
Negative Priming Robustness of the NP effect Variation of objects pictures letters numbers words nonsense shapes... Different tasks identity priming location priming lexical decisions stroop tasks... Reaction determination key pressure voice recording... D C D A D B C match BALL mismatch
Negative Priming Oddities of Negative Priming 4 Time Course NP even after a month paradoxical effects difference of reaction time [ms] 3 2 - -2 5 ms ms 5 ms NP2 NP PP PP2-3 response stimulus interval (RSI) [ms] Presence of Distractors no NP if a trial obviously requires no selection NP if this is not predictable Influence of Strategies while comparison sig. NP if the perceptual order is fixed no NP if both objects in focus
Negative Priming Theories A Few Explanations of Negative Priming Distractor Inhibition distractor representations are actively inhibited inhibitory rebound pushes distractors below baseline neural rate code model of distributed representations Houghton, G. and Tipper, S. P. (994). A dynamic model of selective attention. In Dagenbach, D. and Carr, T., editors, Inhibitory mechanism in attention, memory and language, 53 2, Orlando, FL. Academic Press. Adaptive Thresholding adaptive threshold for action decision forced decay of representation strength while conflict phenomenological model of representation activation Schrobsdorff, H., Ihrke, M., Behrendt, J., Hasselhorn and M., Herrmann, J. M. (27). A Computational Approach to Negative Priming. Connection Science, 9(3), 23-22.
Negative Priming Theories A Few Explanations ctd. Episodic Retrieval response generation by direct computation vs. retrieval automatic retrieval of former episodes by similarity a change in response to similar stimuli causes NP Rothermund, K.; Wentura, D. & Houwer, J. D. (25) Retrieval of incidental stimulus-response associations as a source of negative priming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 3, 482-495 Dual Mechanisms inhibition as well as retrieval processes are present direct computation by distractor inhibition depending on the paradigm, they contribute more or less May, C. P., Kane, M. J., and Hasher, L. (995). Determinants of negative priming. Psychological Bulletin, 8():35 54.
Negative Priming Theories Unifying all Theories How to concretize the theories for comparison? Construct and implement a computational model...... to quantify, integrate and cross-test negative priming theories. Main Problems Most theories give no quantification Many different components are necessary Integration of different paradigms Cope with inconsistencies between theories
.4.2.8.6.4.2.5 2 2.5 3 x 4 8 x 3 7 6 5 4 3 2.9.8.7.6.5.4.3.2. x 3.5 2 2.5 3 x 4.5 2 2.5 3 x 4.4.2.8.6.4.2.5 2 2.5 3 x 4.9.8.7.6.5.4.3.2..9.8.7.6.5.4.3.2..5 2 2.5 3 x 4.5 2 2.5 3 x 4.9.8.7.6.5.4.3.2..5 2 2.5 3 x 4 The General Model The General Model color Perceptual Input shape Bus Feature Layers word Overview rate code model exponential dynamics adaptive thresholds no synaptic delay target: green task: compare Central Executive Episodic Memory Binding Layer NO Response Action Layer Semantic Layer feature decomposition feature binding semantic representation decaying memory weighted recall action decision
The General Model Feature Layers Background red feature decomposition in the visual pathway green input only to perceived features BALL Ball grey number and topology of feature layers paradigm specific Realization distinct feature layers, here Color, Shape and Word every layer holds one variable for each feature instance feature present input =, otherwise input =.
.4.2.8.6.4.2.5 2 2.5 3 x 4.9.8.7.6.5.4.3.2. x 3.5 2 2.5 3 x 4.4.2.8.6.4.2.5 2 2.5 3 x 4 Unifying Negative Priming Theories The General Model Feature Binding color Binding Layer shape Background the brain has to track object entities features have to be bound together in a flexible way objects are represented by bindings without perception, bindings decay Realization a vector holds indices of feature instance variables the binding has a certain maximum synaptic strength feature instances balance their activation via bindings slow decay of synaptic strength in absence of input Schrobsdorff, H., Herrmann, J. M., and Geisel, T. (27). A feature-binding model with localized excitations. Neurocomputing, 7(-2):76 7.
The General Model Semantic Layer Semantic.8.6.4.2 4 5 6 7 8 9 2 Background paradigms like object naming and comparison rely on semantic classification language evokes semantic representation action initiation by comparing semantic activations to a threshold Realization feature layers with semantic matter project to the semantic layer words and shapes with the same semantic meaning converge a threshold adapting to a global mean divides the representations into sub- and superthreshold for action decision
The General Model Action Layer Action.9.8.7.6.5.4.3.2..6.7.8.9 2 2. 2.2 2.3 x 4 Background only one action at a time possibility to hold action initiation concurrence of different actions Realization input by superthreshold semantic representations additionally activation for no response contributing to threshold action initiation by crossing an adaptive threshold
The General Model Episodic Memory x 3 Episode 9 8 7 6 5 4 3 2 4 5 6 7 8 9 2 Background every finished episode is memorized the memory decays with time similarities between percept and memory trigger a retrieval the more similar the percepts, the stronger the retrieval Realization After a reaction, the entire actual state of the model is stored. A scalar product of percept and memory determines retrieval. Every variable receives the memorized value weighted with the retrieval strength as additional input.
.4.2.8.6.4.2.5 2 2.5 3 x 4 8 x 3 7 6 5 4 3 2.9.8.7.6.5.4.3.2. x 3.5 2 2.5 3 x 4.5 2 2.5 3 x 4.4.2.8.6.4.2.5 2 2.5 3 x 4.9.8.7.6.5.4.3.2..9.8.7.6.5.4.3.2..5 2 2.5 3 x 4.5 2 2.5 3 x 4.9.8.7.6.5.4.3.2..5 2 2.5 3 x 4 The General Model Central Executive color Perceptual Input shape Bus Feature Layers word Background organization of layer interplay has to be adapted to the task by a control instance external to the model target: green task: compare Central Executive Episodic Memory Binding Layer NO Response Action Layer Semantic Layer Realization target color (green) is amplified paradigm specific binding of (semantic) representations to actions
The General Model Simulation NP Sim: Word Picture Comparison.5 Color.5 Shape.5 Word.5.5.5.6.7.8.9 2 2. 2.2 2.3 x 4 3 x Binding.8.6.4.2.6.7.8.9 2 2. 2.2 2.3 x 4.8.6.4.2.6.7.8.9 2 2. 2.2 2.3 x 4 Semantic.6.7.8.9 2 2. 2.2 2.3 x 4.6.7.8.9 2 2. 2.2 2.3 x 4 x 3 Episode Action 8 6 4 2.6.7.8.9 2 2. 2.2 2.3 x 4.8.6.4.2.6.7.8.9 2 2. 2.2 2.3 x 4
Implementation of Theories Houghton n Tipper Essence of the theory The distractor is inhibited. When the input is switched off, persisting inhibition pushes the distractor below baseline. Activation of the former distractor first has to surmount this rebound. Position in the Model feature decomposition rate code model switch between target amplification and distractor inhibition effect of amplification in a specific feature instance But no negative activations..9.8.7.6.5.4.3.2. Color 6 7 8 9 2 3 Houghton, G. and Tipper, S. P. (994). A dynamic model of selective attention. In Dagenbach, D. and Carr, T., editors, Inhibitory mechanism in attention, memory and language, pages 53 2, Orlando, FL. Academic Press.
Implementation of Theories Imago-Semantic Action Model Essence of the theory Decisions about actions are triggered by only one semantic variable being superthreshold. Negative priming is produced by a forced decay of activations if two similar but nonmatching representations clash. Position in the Model adaptive threshold in semantic layer the number and identity of superthreshold variables trigger actions residual bindings produce a flow of activation from perceived objects to complementary variables Semantic.8.6.4.2 4 5 6 7 8 9 2 Schrobsdorff, H., Ihrke, M., Behrendt, J., Hasselhorn and M., Herrmann, J. M. (27). A Computational Approach to Negative Priming. Connection Science, 27.
Implementation of Theories Episodic Retrieval Essence of the theory The onset of a new trial triggers a retrieval of the last episode from memory. Conflicts produce a slowdown in reaction time. Recently: Perceptual similarities trigger mainly a retrieval of the prime response. Position in the Model Episodes are memorized. Perceptual similarity determines retrieval strength. individual retrieval effects to switch between classical and modern episodic retrieval x 3 Episode 9 8 7 6 5 4 3 2 4 5 6 7 8 9 2 Frings, C., Rothermund, K. and Wentura, D. (in press). Distractor repetitions retrieve previous responses to targets. Quarterly Journal of Experimental Psychology.
Implementation of Theories Conclusion Take Home Message We have implemented a general model for selective attention that produces negative priming effects The model concretizes NP theories for direct comparison The model can be applied to all NP paradigms Outlook To account for all theories still requires a lot of reading Determination of theory-specific set-screws inclusion of aging effects application to perception based action selection in robotics
Implementation of Theories Thanks... C4-Project Mattias Ihrke Jörg Behrendt Marcus Hasselhorn J. Michael Herrmann NP-Theorists Björn Kabisch Christian Frings Aging Shu-Chen Li Timo von Oertzen EEG Experiments Torsten Wüstenberg Henning Gibbons Ralph Meier Miguel Valencia Ustárroz... and to You!