1 of 47

Informative subject line
Address the email formally (use the correct prefix/title when addressing the professor)
Be direct and straight to the point (don’t beat around the bush)

If you just want to talk let them know
If you want a position tell them you’re interested

Show that you put in the time to research the chosen field

ie: tell them that you found their ___ paper interesting

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

15 of 47

How NOT to Cold Email…

Overly casual, lacking formality

Use of incorrect font

Inconsistent grammar
Shows no forethought or care
Gives the professor no background information

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

16 of 47

How not to email prospective grad school advisors

Decomposition.ai blog post

By Lindsey Kuper

https://decomposition.al/blog/2020/11/25/how-not-to-email-prospective-grad-school-advisors/

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

17 of 47

How to Cold Email!

Informative and formal address and header
Gives background on who you are and why you are good fit for research
Shows you are passionate about the field
Goes out of the way to make a meeting

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

18 of 47

Another Good Example (High School Student)

Hello Professor {name},

My name is {name} and I am a junior at {high school} in {city}, California. I came across your page while browsing the UCSC Baskin Engineering website for potential research opportunities. In the process of trying to figure out what field I would like to focus on in the future, I like to seize every available opportunity to learn more about the fields I am curious about, and I have started to look for ways to involve myself in research throughout the year.

I would like to ask if any postdoc or graduate students in the {field} and {subject} group would be interested in having a high school research assistant/intern to help out remotely with any simple tasks (tabulating data, writing simple scripts, testing, etc) that a high schooler could possibly do. I would be thrilled to apply the skills I have been learning to try to solve real-world problems, and I am willing to do any tasks necessary.

Regarding my experience, I have taken courses in machine learning, Java, and Python. I have also completed {CS course} and {CS course}. I’ve also completed a project with {AI group} where my group and I analyzed stock reviews and utilized NLP to predict stock prices. Additionally, I am a robotics team member and have basic experience in assembling hardware parts and robotic designs. I have linked my resume and my unofficial high school transcript for more relevant information. I would appreciate the opportunity to help you and I believe this experience would be invaluable to helping me decide my path through college and beyond. Thank you for your valuable time and I look forward to hearing back from you!

Best regards,

{name}

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

19 of 47

Graduate Schooling + Timeline

Relevance

If you plan on pursuing a PhD or a M.S. degree with research, it is necessary to pursue research in your undergrad.
If you plan on pursuing a PhD, you must have a paper published or under review by the time you are applying to grad school.
Having a paper published in undergrad will drastically increase your chances of getting into a good graduate research program.

Top schools will typically only take students who have published papers. c

Timeline

Freshman/Sophomore year - get involved in a research lab!

Sophomore/Junior year - make significant contributions to a project, start working towards a paper

Winter of Junior year - submit your paper to journals and conferences

By fall of senior year - Have your paper submitted to a conference/journal or published , list your paper in your graduate school application.

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

20 of 47

My Experience in Research

A short talk by Coen

Slides: tinyurl.com/acmResearchPresentation

21 of 47

Where it all began: CSE 30

Coen Adler

I’m doing research on Computational Ecology anyone want to help?

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

22 of 47

CSE 30 Discussion Section

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

23 of 47

Professors

Leilani Gilpin

Artificial Intelligence Explainability Accountability (AIEA) Lab

Luca de Alfaro

Computational Ecology

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

24 of 47

Diffusion Models for Species-Specific Functional Habitat Connectivity

University of California, Santa Cruz

Team members:

Luca de Alfaro, Natalia Ocampo-Peñuela, Coen Adler, Artie Nazarov, Jasmine Tai, Natalie Valett

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

25 of 47

Habitat Connectivity

Habitat connectivity maps species ability to traverse over a landscape
It is essential for viability of a species

Larger habitats
Greater genetic diversity

How do we map it?

Habitat Connectivity computed for Acorn Woodpecker

Image: Birdsoftheworld.org

Slides: tinyurl.com/acmResearchPresentation

26 of 47

Project Breakdown

Correlation

Web Application

EcoScape

Geospatial Data

SCGT

(UC Santa Cruz Geographical Toolkit)

Data Collection

Habitat/Landscape generation

Connectivity

Algorithm

Stochastic Diffusion

Optimization

Iteration/GPU parallelization

Artifacting

Validation

eBird Validation

Hop Size vs Dispersal Distance

Prototyping

Google Cloud

Google Earth Engine

Google Maps API

Slides: tinyurl.com/acmResearchPresentation

27 of 47

Project Breakdown

Correlation

Web Application

EcoScape

Geospatial Data

SCGT

(UC Santa Cruz Geographical Toolkit)

Data Collection

Habitat/Landscape generation

Connectivity

Algorithm

Stochastic Diffusion

Optimization

Iteration/GPU parallelization

Artifacting

Validation

eBird Validation

Hop Size vs Dispersal Distance

Prototyping

Google Cloud

Google Earth Engine

Google Maps API

Slides: tinyurl.com/acmResearchPresentation

28 of 47

Connectivity Algorithm

Input

Result

Habitat

Landscape

(Resistance)

Repopulation

Gradient/Flow

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

29 of 47

Connectivity Algorithm

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

30 of 47

Mapping Connectivity

Existing models limited by efficiency
Our model utilizes a pixel-level diffusion algorithm

Utilizes parallel GPU computation
Up to 1,000 times faster

Measuring accuracy via correlation with citizen-science data from eBird

First quantitative validation technique using citizen-science data
On par with leading models

Habitat Connectivity/Flow computed for Acorn Woodpecker

Slides: tinyurl.com/acmResearchPresentation

Leading connectivity models used today are unfortunately very slow due to connectivity being so computationally intensive, resulting in very high costs in both time and money to run these models.

These costs only increase in proportion to the geographic size and the number of species being ran, making truly informed conservation decisions extremely costly and often infeasible.

Here we present EcoScape, a novel connectivity approach which simulates species dispersal via a diffusion process, offering properties that are able to be parallelized on GPU’s for computation up to 1000x faster to the leading models today.

We’ve also developed the first quantitative validation technique for connectivity models, which leverages publicly available citizen-science data from eBird to measure correlation between connectivity from our and other models with field observations. Our model has a correlation that’s on par with leading models, while running in just a fraction of the time.

31 of 47

Mapping Flow

Generates a Flow layer

Flow indicates a pixel’s impact on connectivity

Higher flow equates to higher “foot” traffic

Used to find important corridors in connectivity

Calculated using backpropagation via diffusion

Habitat Flow computed for Acorn Woodpecker

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

The current most-used connectivity models use circuit-based approaches. Even the best of these are very slow to compute, and therefore not scalable to large geographic scales.

Our Model, which we’ve coined EcoScape, proposes a novel approach to simulating species dispersal via a diffusion process, offering properties that are able to be parallelized on GPU’s for much faster computation.

Our model runs up to 1000x faster to the leading models today, while providing just as accurate predictions of species dispersal and connectivity.

Due to the ML-based approach being taken, we’ve discovered a way to easily compute a ‘flow’ layer in addition to the connectivity. This flow layer indicates a pixel’s importance in facilitating dispersal, and therefore is a very useful indicator of what land should be protected and restored.

32 of 47

Geo For Good 2022/2023

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

33 of 47

Artificial Intelligence Explainability Accountability (AIEA) Lab

Lab members:

Leilani Gilpin, Coen Adler, Shengjie Xu, Oliver Chang, Olivia Wang, Li Liu, etc…

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

34 of 47

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

35 of 47

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

36 of 47

Introduction to GANs

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

37 of 47

Post-Hoc Explanation Disagreements

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

38 of 47

Measuring Disagreements

Discord: tinyurl.com/acmDiscordLink

39 of 47

Fair Decision Trees

Discord: tinyurl.com/acmDiscordLink

40 of 47

Fair Augmentation of Decision Trees

41 of 47

Example

42 of 47

Baylearn 2023

43 of 47

ACM Undergraduate AI Research Lab

Lab members:

Coen Adler, Arnav Kartikeya, Kevin Bachelor, Aled dela Cruz, Ihina Nath, Joshua Schuman, etc…

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

44 of 47

Utilizing Segment Anything Model for Assessing Localization of GRAD-CAM in Medical Imaging

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

45 of 47

Analysis of Training PINNs on Multiple Computational Fluid Dynamics Models Consisting of Different Shapes

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

46 of 47

Any Questions?

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink

47 of 47

Important Links

Slides: tinyurl.com/acmResearchPresentation

Discord: tinyurl.com/acmDiscordLink