Yifan Gao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yifan Gao has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Abhishek Datta
Research and Development, Soterix Medical Inc., Woodbridge, NJ, USA
2 shared publications - Dennis Quangvinh Truong
Research and Development, Soterix Medical
2 shared publications Jeffrey Moreno
2 shared publicationsKamran Nazim
2 shared publications- Rashel Mejia
Columbia University
2 shared publications - Yishai Valter
Research and Development, Soterix Medical Inc., Woodbridge, NJ, USA
2 shared publications
Latest Publications
- Open Access
- 5 June 2026
- Published by Elsevier in Transcranial Magnetic Stimulation
- Vol. 7
Virtual neuronavigation for parcel-guided TMS
- 5 June 2026
- Published by Elsevier in Transcranial Magnetic Stimulation
- Vol. 7
Open AccessVirtual neuronavigation for parcel-guided TMS
- Open Access
- 12 May 2026
- Published by Wiley in Psychophysiology
- Vol. 63 (5) , e70308
The Reward Positivity Tracks Positive Reward Prediction Errors From Feedback to Cues During Reinforcement Learning
How does the brain learn to predict rewards? According to temporal difference (TD) learning theory, reward prediction errors (RPEs) should shift from the time of outcome delivery to earlier predictive cues as stimulus‐action‐outcome associations are learned. The reward positivity, an electrophysiological signal believed to index sensitivity of the anterior midcingulate cortex to positive RPEs, should progressively transfer from feedback to predictive cues during learning. However, this core prediction of the reward positivity has remained largely untested. We recorded the EEG from 73 healthy adults performing a probabilistic selection task (PST) with extended training trials. The reward positivity amplitude was measured at both feedback and cue presentation during early and late training phases (first and second halves, respectively). To examine individual differences in RPE‐related learning processes, we split participants into rapid and slow learner groups and fit Q‐learning models to estimate separate learning rates for positive and negative feedback. Results showed clear evidence of temporal backpropagation: early in learning, the reward positivity appeared when feedback was delivered, but late in learning it disappeared from feedback and instead emerged when predictive cues appeared. Rapid learners showed more pronounced shifts in reward positivity from feedback to cue, consistent with their higher learning rates. These findings provide the first clear evidence that reward positivity demonstrates the temporal backpropagation predicted by TD learning theory. The results validate reward positivity as a neural marker of positive RPEs and highlight the importance of examining both cue‐related and feedback‐related brain responses to fully characterize reinforcement learning processes. Our findings have important implications for understanding individual differences in reinforcement learning and for interpreting the reward positivity in clinical populations and across the lifespan.
- 12 May 2026
- Published by Wiley in Psychophysiology
- Vol. 63 (5) , e70308
Open AccessThe Reward Positivity Tracks Positive Reward Prediction Errors From Feedback to Cues During Reinforcement Learning
How does the brain learn to predict rewards? According to temporal difference (TD) learning theory, reward prediction errors (RPEs) should shift from the time of outcome delivery to earlier predictive cues as stimulus‐action‐outcome associations are learned. The reward positivity, an electrophysiological signal believed to index sensitivity of the anterior midcingulate cortex to positive RPEs, should progressively transfer from feedback to predictive cues during learning. However, this core prediction of the reward positivity has remained largely untested. We recorded the EEG from 73 healthy adults performing a probabilistic selection task (PST) with extended training trials. The reward positivity amplitude was measured at both feedback and cue presentation during early and late training phases (first and second halves, respectively). To examine individual differences in RPE‐related learning processes, we split participants into rapid and slow learner groups and fit Q‐learning models to estimate separate learning rates for positive and negative feedback. Results showed clear evidence of temporal backpropagation: early in learning, the reward positivity appeared when feedback was delivered, but late in learning it disappeared from feedback and instead emerged when predictive cues appeared. Rapid learners showed more pronounced shifts in reward positivity from feedback to cue, consistent with their higher learning rates. These findings provide the first clear evidence that reward positivity demonstrates the temporal backpropagation predicted by TD learning theory. The results validate reward positivity as a neural marker of positive RPEs and highlight the importance of examining both cue‐related and feedback‐related brain responses to fully characterize reinforcement learning processes. Our findings have important implications for understanding individual differences in reinforcement learning and for interpreting the reward positivity in clinical populations and across the lifespan.
- Open Access
Virtual Neuronavigation for Parcel-guided TMS
Background: rTMS for depression is partially effective which is likely in part due to the treatment coil positioning being based on scalp-based approaches. Such an approach results in not accounting for individual anatomy and therefore delivering electrical currents to non-optimal brain targets. In a pilot study, we demonstrated that rTMS targeted to parcel 46 (p46) of the HCP atlas led to 100% response in patients who were resistant to the standard TMS. Implementing parcel-guided targeting requires neuronavigation. However, neuronavigation use in clinical practice is limited. Besides added cost, need for individual MRI, leveraging new targeting advances requires expertise in 3D brain reconstruction and processing. Methods: We sought to develop a virtual neuronavigation system in which MRI images are processed on the backend to determine p46 and a corresponding individualized 3D printed scalp headgear is developed. The headgear includes markings for the location corresponding to p46 and other anatomical guides. Headgear development involved an iterative sequence of 3D CAD modeling, printing, and testing to meet the gold standard neuronavigation-guided location. Individual headgear accuracy is then verified by positioning on the intended subject and comparing alignment with the location on the scalp and brain intersecting the neuronavigation beam targeting p46. Reproducibility testing was also performed. Results: Across 16 test subjects on which the final version of the headgear could be tested, individualized headgear was found to be 8.31± 4.93 mm from the neuronavigation-determined scalp location. The corresponding distances on the brain was 7.94± 4.99 mm. Within-subject reproducibility (against centroid formed by 3 measurements) was 1.21 ± 0.84 mm across n= 8 subjects. Conclusion: Individualized headgear targeting presents a viable option for practices to leverage state-of-the-art targeting advances without investing in neuronavigation hardware.
Open AccessVirtual Neuronavigation for Parcel-guided TMS
Background: rTMS for depression is partially effective which is likely in part due to the treatment coil positioning being based on scalp-based approaches. Such an approach results in not accounting for individual anatomy and therefore delivering electrical currents to non-optimal brain targets. In a pilot study, we demonstrated that rTMS targeted to parcel 46 (p46) of the HCP atlas led to 100% response in patients who were resistant to the standard TMS. Implementing parcel-guided targeting requires neuronavigation. However, neuronavigation use in clinical practice is limited. Besides added cost, need for individual MRI, leveraging new targeting advances requires expertise in 3D brain reconstruction and processing. Methods: We sought to develop a virtual neuronavigation system in which MRI images are processed on the backend to determine p46 and a corresponding individualized 3D printed scalp headgear is developed. The headgear includes markings for the location corresponding to p46 and other anatomical guides. Headgear development involved an iterative sequence of 3D CAD modeling, printing, and testing to meet the gold standard neuronavigation-guided location. Individual headgear accuracy is then verified by positioning on the intended subject and comparing alignment with the location on the scalp and brain intersecting the neuronavigation beam targeting p46. Reproducibility testing was also performed. Results: Across 16 test subjects on which the final version of the headgear could be tested, individualized headgear was found to be 8.31± 4.93 mm from the neuronavigation-determined scalp location. The corresponding distances on the brain was 7.94± 4.99 mm. Within-subject reproducibility (against centroid formed by 3 measurements) was 1.21 ± 0.84 mm across n= 8 subjects. Conclusion: Individualized headgear targeting presents a viable option for practices to leverage state-of-the-art targeting advances without investing in neuronavigation hardware.