top of page

國科會​傑出研究獎及吳大猷先生紀念獎演講

BCKUO.jpg

​ 國科會傑出研究獎 

 

郭柏呈 副教授

郭教授研究專長為認知與知覺心理學,研究興趣為探討選擇性注意在知覺、工作記憶、長期記憶以及靈活反應選擇中的角色。郭教授曾獲吳大猷先生紀念獎,主持多項國科會計畫,現為國立臺灣大學心理學系副教授。

郭柏呈老師

Maintaining with the Benefit of Expectation

10/26(六)

Working memory (WM) and attention often work together in a mutually supportive manner to guide flexible and adaptive behaviours. Because WM is highly limited in capacity, attention plays an important role in anticipating and gating information that is most relevant to behavioural expectations. In turn, WM controls attention by maintaining task goals, allowing attention to be directed towards items that match these goals. In this talk, I will present recent work from my lab on the interaction between WM and attention. Using EEG, MEG, and fMRI, we focus on the neural mechanisms underlying this interaction, and how it guides human behaviours. In the first section, I will present evidence showing how electrophysiological activity tracks content-specific WM capacity during the retention interval of WM. In the second section, I will demonstrate how temporal expectation based on duration variability can modulate the neural dynamics of alpha oscillations that precede the onset of the memory test. In the final section, I will introduce a MEG-fMRI fusion approach and explain how we apply this method to investigate spatiotemporal neural dynamics for top-down modulation of category-specific information. Together, these studies provide novel evidence for the anticipatory, adaptive nature, and flexibility of our mind and brain.

PHH.jpg

​ 吳大猷先生紀念獎 

黃柏僩 副教授

黃教授研究專長涵蓋:心理計量學、統計建模、機器學習,研究興趣為結構方程模型與機器學習之應用。黃教授主持多項國科會計畫,現為國立政治大學心理學系副教授。

黃柏僩

特徵重要性於機器學習之統計推論

10/26(六)

近年來,機器學習(machine learning,簡稱ML)在電腦視覺與自然語言處理等科學領域帶來了革命性的影響。儘管在心理學領域,研究者仍偏好使用線性模型來檢驗心理學理論,但ML確實為實徵研究者提供了另一種詮釋資料之可能性。典型ML算則之運作常被認為是一黑盒子(black-box),難以對其進行詮釋與推論。近期可解釋ML之發展,使得研究者得以窺視黑盒子內部之樣貌,而檢驗特徵重要性(feature importance,簡稱FI)之統計推論程序,更令透過ML算則來檢驗實質理論(substantive theory)一事變得可能。本演講旨在比較幾種檢驗FI之統計推論程序,包括殘差排序檢定(residual permutation test,RPT)、條件預測影響(conditional predictive impact,CPI)、移除單一共變量(leave-one-covariate-out,LOCO)、以及嵌入估計(plug-in estimation,PIE),我們將說明這些方法之建構原則與試圖推論之對象,並透過模擬研究來評估這些方法之實徵表現,盼望能給心理學領域之ML應用者一些實務上之指引。

bottom of page