Dr Yi-Shin Lin

Profile

I joined the Institute for Transport Studies in Aprial 2020. I received my PhD (2015) and MA (2007) in Experimental Psychology and Cognitive Neuroscience from the University of Birmingham and the City University of New York. I studied in Psychology for BSc in the National Taiwan University (2002).

Responsibilities

• Mathematical modeller in Human and AI cognition in COMMOTIONS project

Research interests

My main interests are in human vision, decision-making, Bayesian computation, and computational modelling. I began my PhD project with an interest in examining the process of visual search, and designed a computational model to analyse the entire spectrum of distributions in two-choice RT data. I than taught myself the basics of Bayesian modelling and used that knowledge to develop a new computational model, using the hierarchical Weibull distribution to fit the RT data from a series of visual search experiments. This work contributes to the field, by showing that one can infer the processing of visual search by examining different bands (e.g., fast vs. slow RTs) in the RT distributions. I published this software on GitHub and documented it in the paper “Modeling visual search using three-parameter probability functions in a hierarchical Bayesian framework” (2015) when the open science and data movement was still in its early time. 

During my time with the Tasmanian Cognition Laboratory, I led two projects developing new statistical methods using Bayesian computation. In the first project I designed a new version of the complex decision-making model, Piecewise Linear Ballistic Accumulator model. This version harnesses parallel heterogenous computation using GPU-based CUDA and the C++ language. I also created a tool to allow fitting complex decision-making models which were previously unfeasible. I have shared this work with other researchers via GitHub and documented its details in the paper, “Parallel probability density approximation” (2018). My second lead project was a generic solution for conducting hierarchical Bayesian modeling. I implemented a Bayesian sampling method, population-based Markov Chain Monte Carlo (MCMC), which is particularly effective in cognitive models featuring correlated parameters. The source code is available on my GitHub. This work was summarised in a paper, titled  “Evidence accumulation models with R: A practical guide to hierarchical Bayesian methods”.

My current research conducting in the ITS is to apply the above-mentioned skills and knowledge in studying the road user interactions under the COMMOTIONS project, which is led by Prof. Markkula. Our efforts in the last year (2020) found that the lane-merging interactive behaviours between drivers can be accounted for by a machined-learned model, by applying the Long Short-Term Memory (LSTM), recurrent neural-network (Srinivasan et al, 2021). This finding goes hand-in-hand together with our new cognitive model that applying the basic principle of utility maximisation, discrete motor primitives, (sensory) evidence accumulation to successfully human travelling trajectory and decisions happening in-between when pedestrians are interacting with drivers (Lin et al., 2021).  

A selection of the projects I have contributed with their data and codes in the OSF.

1. https://osf.io/av3cm/
2. https://osf.io/p4pdh/
3. https://osf.io/pbwx8/