Tyler Malloy

Tyler Malloy

They/Them

Postdoctoral Researcher

PhD Defense

5 minute read

Date:

It has been a while since I made a blog post, mostly becase I have been working hard on my PhD Dissertation and practicing for my defense, which you can watch a practice run through of at this link. Since the last time I made a post I have defended my dissertation and I am now in the final stages of my work at Rensselaer. I have had an amazing time working at RPI with my mentor Chris R. Sims, as well as working in collaboration with the Reinforcement Learning research group at IBM, particularly with my mentor Tim Klinger. The next big step for me is starting my postdoctoral fellowship research position working with Coty Gonzalez at Carnegie Mellon in December of this year.

So onto the dissertation defense itself. I had practiced the presentation around a dozen or so times in the leadup to the day of the defense, and I felt that I had I good handle on the flow of the presentation. One thing that was difficult, that I assume many PhD defense presenters find, is that discussing a 100 page dissertation in 45 minutes can be very challangeing. I was lucky partially in that I had given an approximately 30 minute presentation to the Cognitive Science department at RPI only a few weeks prior, as a part of a yearly requrement for graduate students in the department.

That earlier presentation ended up being a preview of what I would discuss in my dissertation defense, and was very helpful in finding out what I needed to describe to someone who was unfamiliar with my disseration topic. Typically for a broad audience, like a university department, it makes sense to eschew any highly technical aspects such as loss functions used for training, or algorithms that define utility in decision making. This was my approach to the earlier introductory presentation, and I felt that a high level description of the real-world psychological phenomenon I was interested in was useful for that.

When it came time to begin putting together my dissertation defense, I started with this preliminary presentation, and added on any necessary detail that I would think someone in my dissertation committee, or another researcher familiar with this area, would be wondering. For me specifically, this was information on how my model was trained to predict human behaviour, and how the comparison models made their own predictions of human behaviour. This allowed me to compare in more detail my proposed model and related methods for explaining behaviour, to better explain where the differences were coming from.

The most significant aspect of my model that I was able to explain in more detail in the longer version of my presentation was that it is able to reflect biases present in human decision making as a result of information constraints, and not through the intoduction of additional parameters that alternative models rely on. I think that this is the easiest to understand benefit of my proposed model, and I was happy that I got to explain it in my dissertation defense. Though in order to quantify that difference, I needed to explain the 2 alternative versions of predicting human behaviour, one that does so without explicitly modelling the observed bias and one that does, and this was required for the two types of task in my experiment, as well as 2 alternative deep neural network based methods for a total of 6 comparison models.

Having this much more to discuss and only a slightly longer presentation time meant cutting or summarizing a lot of the content from my earlier presentation, but deciding what to cut and what to keep was easier than expected. There were a few long descriptions of real-world examples in my first presentaiton that I significantly summarized. Additionally, explaining the background of the model I proposed was shortened, given that the main audience would have (hopefully) read my dissertation. None of these things should totally be removed from an ideal dissertation defense, though explaining everything for a general audience is also not totally necessary.

With those changes to the presentation, and the previously mentioned dozen or so practice runs, I was able to succesfully defend my dissertation with a few minor revisons requested by my committee. During the presentation itself there were some clarifying questions from my committee that weren’t too difficult to adreess. Afterwards the only question form the general audience came from my friend who works in vision science, wondering about the broad implications of my work for her field. I would have liked to make some strong claims about visual representations, but was more focused on learning and didn’t want to make any claims not fully backed up by the experiment I presented.

After a short deliberation period I was welcomed back into the room by my advisor Chris, who had a smile on his face as he said the words I had been waiting nearly 5 years to hear “Congratulations Dr. Malloy”.

You May Also Enjoy

Dynamic Decision Making Lab Yearly Review

2 minute read

Published:

The Dynamic Decision Making Laboratory has issued it’s yearly review for 2024 which summarizes the work from the lab that occured during the past year. It is available online here. I have copied below my own section of the yearly review, and encourage people interested in my work to check out the annual summary. This will also be of interest to those who would like to get an idea of the work that goes into running a relatively large research lab in decision sciences.

Descisions From Description and Experience

8 minute read

Published:

One of the interesting lines of recearch that I have recently become more involved with is the distinction between making decisions based on description and making them based on experience. In this research, these are referred to as Decisions from Experience (DfE) and Decisions from Description (DfD). In the past, I have studied differences and similarities between learning and decision making, which is as similar area of research in cognitive psychology. In the future I will be exploring whether these differences between decisions made from description and experience also occur in large language models, but for this post I will be focusing on the background of these differences.

Measures of Similarity

6 minute read

Published:

This post is inspired by the recent preprint that I released called ‘Leveraging a Cognitive Model to Measure Subjective Similarity of Human and GPT-4 Written Content’. In this work, we develop a method for estimating human’s subjective similarity using a cognitive model. Similarity between stimuli has been a major theme in my research, and it is a major challenge to cognitive modeling in general. In this post I will detail the importance of metrics of similarity that accurately reflect how humans perceive the similarity of stimuli they are presented with. I will focus on Instance Based Learning models, and in a later post discuss how similarity is relevant for Reinforcement Learning models and how we apply Generative AI representations onto determining similarity.

Multi-University Research Initiative

26 minute read

Published:

This blog post contains my unedited notes that I took while attending the Multi-University Research Initiative Program Review this year. This was an interesting meeting where I presented my work in cognitive models applied onto large languge model to improve their usefullness in educational settings, specifically for phishing email education.