Anne Collins, Professor

Closed (1) Learning and generalization

Closed. This professor is continuing with Fall 2018 apprentices on this project; no new apprentices needed for Spring 2019.

To learn quickly in a new situation, we often simply figure out the best way to reuse skills we already learned in the past. Identifying which skills to generalize is a challenging task. How do we do it?
The proposed research plans to investigate learning and decision making in healthy young adults. In this particular project, we investigate how we learn different skills, and more importantly, how we generalize these learned skills to make learning and exploration faster in novel situations.

This research project involves three steps. First, we will recruit participants and collect data for this experiment. Second, we will analyze the behavioral data to help formulate hypotheses about human learning and generalization. Third, we will do computational model fitting to arbitrate between different models and test which model better explains human data.

Experiments for this project will take place both through RPP and through the Amazon Mechanical Turk platform.

Research assistants will have an opportunity to participate in multiple stages of the research process and to receive mentoring along the way.

In particular, research assistants will be responsible for running the experiment through RPP and Amazon Mechanical Turk, through the several steps required: running the task, monitoring the software and data collection, organizing collected data and analyzing data. They will receive training and supervision in these tasks.

Furthermore, research assistants will be expected to familiarize themselves with the goals and the contents of the project. They should be able to clearly explain the purpose and predictions of the experiment. The mentor will provide guidance in reading some of the relevant literature, to put the specific project in a broader context.

Research assistants will be required to meet regularly with their mentor and to keep track of their progress in the project.

Day-to-day supervisor for this project: Jimmy Xia, Graduate Student

Qualifications: To work on this project, we will use MatLab for RPP, and the Amazon Mechanical Turk platform (https://www.mturk.com/mturk/), with support from Psiturk (https://psiturk.org/). For this project, RAs will need to use code in MatLab, javascript, R, and manipulate SQL databases. Mentors will provide support to learn these skills, but some coding skills are strongly recommended and preferred.

Weekly Hours: to be negotiated

Related website: https://www.ocf.berkeley.edu/~acollins/

Closed (2) Working memory and reward-based learning during adolescence

Applications for Spring 2019 are now closed for this project.

How does working memory develop during adolescence and how does its involvement in learning change?

We know that different parts of the brain mature at different stages during development. The Basal Ganglia, heavily involved in reward-based learning, mature much earlier than the prefrontal cortex, involved in working memory. Correspondingly, previous studies have shown that working memory changes during development - and other studies have shown that reward-based learning changes. But how do the two systems interact with each other? We want to answer this question by testing 7-18 year olds in working-memory and reward-based learning tasks and by creating cognitive models of their performance that can tell us exactly what strategy they used.

We have collected a large developmental sample. Now, we're looking for a research assistant who can perform data analysis in collaboration with us.

Undergraduate researchers will be involved in data collection and data analysis. Specifically, they will be tasked with analyzing data from one of the 5 tasks we administer, an ATARI game called QBert.

Undergraduates will be expected to familiarize themselves with the project, reading suggested literature, so that they can learn more about the hypotheses and goal of the research. They will also be expected to participate in the more general lab experience, with regular group or lab meetings with other lab members and the PI.

Day-to-day supervisor for this project: Sarah Master, Staff Researcher

Qualifications: Applicants should have strong knowledge of and/or interest in neuroscience or cognitive science, and a strong background in computer science or data analysis. Experience with image processing is a plus. Students who have previously worked in or taken a class on computational modeling are encouraged to apply, though experience modeling is not a prerequisite. Students must be motivated, independent, and reliable.

Weekly Hours: to be negotiated

Related website: https://www.ocf.berkeley.edu/~acollins/

Closed (3) Measuring and modeling working memory and reward-based learning

Applications for Spring 2019 are now closed for this project.

Over the course of our lives we learn to make the choices that will maximize reward. As a young adult, you have probably figured out that eating a donut is more rewarding than eating a plain piece of bread. As such, you are more likely to choose to eat the donut. You have also learned more complex behaviors for maximizing reward, like the long-term strategies involved in earning good grades or the praise of your professors. Our question is: How do we learn those complex strategies? Which brain systems contribute to learning from reward, and how do they interact to produce those behaviors? How do these interactions unfold over time?

The proposed research plans to investigate learning and decision making in healthy young adults. We use computerized tasks to probe how multiple systems contribute to let us learn efficiently in different situations; and how these different systems interact with each other. In this particular project, we investigate how rewarding outcomes influence how we act in the future.

This research project is going to focus heavily on data collection, analytics, and computational modeling. This semester we are moving into data collection with functional magnetic resonance imaging (fMRI) so students will learn to collect, pre-process, and work with neural data. The students will work with the mentor to collect and analyze data, and produce mathematical models of decision-making.

First, the apprentice will be involved in experimental design and the writing of experimental code in Matlab/Psychtoolbox for lab testing of human subjects, or in Javascript for online testing of human subjects on Amazon Mechanical Turk. They will also be responsible for the recruitment and testing of human subjects, and will also be involved in analyzing behavioral data and building and testing mathematical models of reinforcement learning and working memory. They will be involved in recruiting and running human subjects in fMRI experiments, and in pre-processing fMRI data.

Undergraduates will be expected to familiarize themselves with the project and read suggested literature so that they can learn more about the hypotheses and goal of the research. They will also be expected to participate in the more general lab experience, with regular group or lab meetings with other lab members and the PI.

Day-to-day supervisor for this project: Sam McDougle, Post-Doc

Qualifications: Applicants should have strong knowledge of and/or interest in neuroscience or cognitive science, and some background in mathematics or computer science. Students who have previously worked in or taken a class on computational modeling are encouraged to apply, though any students with a basic understanding of modeling and/or programming are welcome too. Students must be motivated, organized, and reliable.

Weekly Hours: 9-11 hrs

Related website: https://www.ocf.berkeley.edu/~acollins/

Closed (4) The neural correlates of structure learning

Applications for Spring 2019 are now closed for this project.

Humans are remarkably adept at reusing previously learned knowledge. For example, once you've learned how to use your first Mac computer, it is simple to figure out how to use other Macs by recycling the rules that you've already learned. This process is called generalization. However, using a Windows computer is completely different. The first time you use a Windows computer you will need to learn entirely new rules. And once you understand how to use both Mac and Windows computers, you will need to be able to flexibly switch from one set of rules to the next. This is accomplished by building structure and recognizing contexts.

We know that humans are good at learning rules and when to apply them, but we don't know how that learning is actually accomplished in the brain. This project aims to investigate that question using behavioral testing and electroencephalography (EEG). EEG is a non-invasive way of measuring large-scale brain activity through the scalp. Using EEG, we plan to further understand how humans are able to learn how to apply different rules in different situations, or re-apply rules in situations they've seen before.

Research assistants will have an opportunity to participate in multiple stages of the research process and to receive mentoring along the way. Research assistants will be required to meet regularly with their mentor and to keep track of their progress in the project.

Research assistants on this project will be responsible for data collection, neural data processing, and data analysis. They will be asked to prepare subjects for EEG data collection, run them in the behavioral task, pre-process the EEG data, and may be asked to analyze that data.

Furthermore, research assistants will be expected to familiarize themselves with the goals and the contents of the project. They should be able to clearly explain the purpose and predictions of the experiment. The mentor will provide guidance in reading some of the relevant literature, to put the specific project in a broader context.

Day-to-day supervisor for this project: Sarah Master, Staff Researcher

Qualifications: Applicants should be interested in cognitive neuroscience and willing to spend 8-10 hours a week on this project. Some experience testing human subjects is helpful, but previous experience is not a must. We are looking for hard-working, motivated, and passionate students. If you have any experience with coding or computational modeling, please mention it in your application.

Weekly Hours: 9-11 hrs

Related website: https://www.ocf.berkeley.edu/~acollins/