Using RealTime Computational Modeling to Individually Optimize Tone

Using Real-Time Computational Modeling to Individually Optimize Tone Category Learning 1 1, 2 Seth R. Koslov , Blanco, N. J. , W. Todd Maddox , & Bharath Chandrasekaran 1 The University of Texas at Austin, Psychology; 2 The University of Texas at Austin, Communication Sciences and Disorders Introduction Individualization Uni-Dimensional & Multi-Dimensional models were fit to 40 trial sliding windows, generating a best fit model on each window Ø Learning new speech categories in adulthood is a difficult and complex task, but crucial to learning a new language. Sliding Window: Best Fit Model UD UD CON CON UD SPC Percent Subjects Reaching Criterion SPC 300 Computational Models Uni-Dimensional Height - Reflective 1 Uni-Dimensional Pitch Direction 0. 4 0. 2 0 0 0. 2 0. 4 0. 6 Pitch Height 0. 6 0. 4 T 1 T 2 0 1 T 3 Control 0 0. 2 0. 4 0. 6 Pitch Height 0. 8 T 1 T 2 T 3 T 4 Sub-Optimal Control Optimal Sub-Optimal Conclusions: 0. 8 0. 6 0. 4 0. 2 0. 4 0. 6 Pitch Height 0. 8 T 1 T 2 T 3 1 T 4 0 0 0. 2 0. 4 0. 6 Pitch Height 0. 8 T 1 T 2 1 T 3 Behavioral Results Individualized bootstrap-training, supporting early reflective and late reflexive processing, will lead to enhanced speech category learning Optimal 1 Multi-Dimensional Conjunctive - Reflective 1 0. 2 Hypothesis: 150 0. 5 T 4 Pitch Direction 0. 8 Multi-Dimensional Striatal Pattern Classifier Reflexive 0. 8 0 200 0. 2 1 Ø Can we individualize tone category learning by using real-time computational modeling to predict an optimal “switch” point? 250 0. 7 0. 8 0. 6 0. 8 Direction - Reflective 1 0. 8 Trials ---------------------------------- • Successful early reflective processing enhances performance in reflexive-optimal tasks • The timing of the reflective to reflexive switch is critical in learning • Optimal training involves supporting reflective processing initially and reflexive processing after a “switch-criterion” is reached T 4 • Bootstrap training is more effective than training that focuses on reflexive strategies for speech category learning - More subjects reached a performance criterion - Subjects reached the criterion faster • Provides support for the bootstrap theory of system interaction Future Directions • Future work should look at the effect of differential training on generalization and retention • Good short term learning does not indicate good retention or generalization • Can we improve on the effectiveness of the Optimal condition training parameters? • What neural markers can be used to compute an individually optimized criterion point? All Stimuli 1 Pitch Direction * 1 2 3 4 5 6 7 8 9 10 11 … 38 39 40 41 42 43 44 45 46 47 48 49 However, bootstrap interaction theory 5 within COVIS predicts: Subjects: 98 young adults (18 -28) from the University of Texas community were randomly placed into one of three conditions: 350 1 Percentage Trials: Ø Previous work supports the claim that speech category learning is reflexive-optimal 2, 3, 4 • Specifically: Minimal Feedback > Full Feedback Stimuli: 5 syllables (bu, di, lu, mi, ma) / 4 talkers (2 f, 2 m) / 4 Mandarin tones = 80 natural stimuli Mixed presentation, 6 blocks 480 total trials Trials to Criterion 0. 9 Feedback Matters! Full Feedback: “Correct that was a 2” – Supports reflective processing Minimal Feedback: “Correct” or “Incorrect” – Supports reflexive processing Methods and Design more consistently and more quickly than subjects in the control or sub-optimal conditions Performance Criterion: 1) Model Fit: 55% of the last 20 windows best fit by Multi-Dimensional Models (MD) & 2) Accuracy: 65% correct responses on the last 20 trials Ø Recently, dual-learning systems theory has been applied to the auditory domain: • Derived from the Competition between Verbal and Implicit Systems (COVIS) model of categorization; postulates two learning systems 1: • Reflective – Explicit, rule based learning mediated by the prefrontal cortex; dependent on working memory • Reflexive – Implicit, procedural learning mediated by the dopamine modulation in the striatum; not dependent on working memory Ø Criterion Analysis Results Subjects in the optimal training paradigm reached the criterion 0. 8 0. 6 References T 1 T 2 0. 4 T 3 T 4 0. 2 0 0 0. 2 0. 4 0. 6 0. 8 1 Pitch Height Condition Pre-Criterion Feedback Post-Criterion Feedback N Control Minimal 30 Optimal Full Minimal 36 Sub-Optimal Minimal Full 32 Minimal: Reflexive Full: Reflective • Optimal condition displays an early advantage in performance • When collapsing across initial feedback type: • Full > Minimal (p=. 0016) • Full feedback enhances early learning • Optimal = Control in late performance • Late full feedback inhibits late learning 1) Ashby, F. G. , Paul, E. J. , & Maddox, W. T. (2010). COVIS. E. M. Pothos & A. J. Wills (Eds. ) Formal Approaches in Categorization. Cambridge University Press, New York. 2) Chandrasekaran, B. , Koslov, S. , Maddox, W. T. (2014). Toward A Dual-Learning Systems Model of Speech Category Learning. Frontiers in Psychology. 5(825), 1 -17. 3) Chandrasekaran, B. , Yi, H. , Maddox, W. T. (2014). Dual-learning systems during speech category learning. Psychonomic Bulletin and Review. 21, 488495. 4) Maddox, W. T. , Chandrasekaran, B. , Smayda, K. , Yi, H. , Koslov, S. , Beevers, C. G. (2014). Elevated depressive symptoms enhance reflexive but not reflective auditory category learning. Cortex. 58, 186 -198. 5) Paul, E. J. & Ashby, F. G. (2013). A neurocomputational theory of how explicit learning bootstraps early procedural learning. Frontiers in Computational Neuroscience. 7 (177), 1 -17. 6) Chandrasekaran, B. , Gandour, J. T. , Krishnan, A. (2007). Neuroplasticity in the processing of pitch dimensions: A multidimensional scaling analysis of the mismatch negativity. Restorative Neurology and Neuroscience, 25, 195 -210. Special thanks to Han Gyol-Yi, Yuan Han, Sharon Noh, Jessica Cooper, Kirsten Smayda, and the research team from the Maddox Lab for their help with this project. This work was supported by NIDA Grant R 01 DA 032457 to WTM. For more information, please visit: http: //homepage. psy. utexas. edu/homepage/Group/Maddox. LAB/index. htm or http: //csd. utexas. edu/research/sound-brain-laboratory.