Content I Consume - 1/18/2024

Past weeks editions can be found here  

Hello all,

We’re officially in IAP (Independent Activities Period) and started 2025. I’ve been traveling around while working on research. I spent the first two weeks of January in LA visiting my girlfriend and now I’m making my way back to Boston. My first stop is Madison, Wisconsin, where I am now. Next a quick three night stint in NYC and then I’ll be back in Boston for the rest of IAP.  

Physics

Podcast from Notebook LM: Product Manifold Machine Learning for Physics V1

Linked above is a 17 min AI podcast explaining my most recent paper. It provides a great overview of the paper and showcases the power of these new AI tools in expanding accessibility to high level and complex topics.

Notebook LM is an exciting new tool develoved by Google. So far, I’ve mainly been experimenting with their audio summary tool which generates a two-person podcast describing a PDF you have uploaded. The method works remarkably well and can explain complex topics well. I’ve been particularly excited about Notebook LM as a way to further explain my research to a non-expert audience.

The AI overview isn’t perfect. In my opinion, the AI overview aggrandizes the results. It does a nice job of explaining the topic, but doesn’t understand the implications of the work well.

Content I Consume - 12/30/2024

Past weeks editions can be found here 

Hello all,

My senior fall semester is officially over and all 10 of my PhD applications are submitted! Even after returning home to North Carolina, it’s hard to leave behind the level of stress these couple of months have normalized.

At MIT, we have the month of January effectively off, so I’m excited to have some time to decompress and focus fully on new research projects.

Physics

Paper: Product Manifold Machine Learning for Physics V1 by Nate S. Woodward et al.

I began this project my sophomore year and I’m excited to finally have it publically available.

In this paper, we explore new data representations for hierarchical datasets in physics. Hierarchies in nature are often formed by processes that evolve through a splitting. For example, a tree grows originally from the trunk but successively splits into smaller and smaller branches. In this evolution, there is a clear hierarchy with the trunk as the most fundamental aspect with each splitting one level lower in the hierarchy.

As you might imagine, for datasets formed through hierarchical processes finding a way to represent our data in a model that also optimally represents the hierarchical structure is key to capturing all the available features of the dataset.

To accomplish this, we leverage non-Euclidean manifolds. In our case, these are curved surfaces. It’s been proven mathematically that some non-Euclidean surfaces can optimally represent the hierarchical structure. In our approach, instead of fully representing data in non-Euclidean manifolds, we consider a combination of several representations on several manifolds, forming a Product Manifold. Specifically, Product Manifolds are Cartesian products of constant curvature Riemannian manifolds (for this paper, we only consider hyperbolic, Euclidean, and spherical geometries).

We develop MLP and transformer models employing product manifold representations and benchmark their performance on particle jets. Read the paper to see why particles are hierarchical objects and how our models perform!

Miscellaneous

Podcast: Beauty in Knowledge

For my final project Producing Podcasts in the spring of my Junior year, I made a 12 minute podcast exploring how learning more about a subject changes our preception. This topic was motivated by questions such as, does the muscian apprechaite music more before or after they are trained in the theory and structures? To me this was never an obvious question and something I’ve discussed before in a blog post here.

In this podcast, I interview three people in different fields at different stages of mastery. I think it’s a great exploration of the topic with some interesting people and all at about 12 min.

Content I Consume - 11/14/2024

Past weeks editions can be found here 

Hello all,

Boston is growing cold and the sun is setting at 4PM. In some ways, this feels more normal than the 70 degree days we’ve had over the past weeks. Life is moving fast with 1 month + 1 day left until I submit my PhD applications. It’s great to see everything come together, but the uncertainty of not knowing where I’ll be for the next 5+ years can be draining.

Today marks the two-year anniversary of my cancer surgery, what I consider as two-years cancer free. I feel so grateful for the friends, new experiences, and the life I am so privileged to live.

While I’m still grappling with the mental challenges of being confronted with my mortality, I’ve come so far in these two years. I couldn’t see myself getting through my diagnosis without the support of my family, friends, and my amazing partner. I am forever grateful.

Physics

Textbook: Scattering Amplitudes by Henriette Elvang and Yu-tin Huang

In high energy physics, scattering amplitudes describe processes of fundamental particles interacting. If you want to understand how likely it is for two photons (light particles) to hit each other and produce a quark and an anti-quark, you would compute the scattering amplitude for this “process.” Where the process would be described as photon + photon to quark + anti-quark.

Study these fundamental interactions through the formalism of scattering amplitudes is a common way to make predictions in theoretical physics about fundamental particle interactions. For example, say I do some mathematics and calculate the scattering amplitude for a+b to c+d (where a,b,c,d are particles). If I go and look at experimental data and see ‘c’ and ‘d’ particles being produced too much or too infrequently, then I know my theory doesn’t align with the experimental results.

This discrepancy can be resolved in a number of ways:

1. Theory is wrong: Are we making mistakes in the calculations? Are there new particles that we aren’t accounting for?

2. Experiment is wrong: Are we biasing our dataset? Are we accounting for all experimental uncertainties?

We have a robust and powerful theory of three of the four fundamental interactions in the standard model of particle physics. However, it is quite difficult to actually calculate to high precision what the standard model predicts. This is the challenge of theoretical physicists and scattering amplitudes are one of the tools they employ.

Miscellaneous

Video: Watch these 8 minutes to live longer - Bryan Johnson

Bryan Johnson is multi-millionaire entrepreneur who now focuses his entire life on improving human longevity. I have become a fan of him recently. I think he’s likely controversial and exaggerates some claims (saying he’ll live to 200 years old), but I think he’s doing some exciting things.

To push forward our understanding of longevity he experiments on himself to reduce his rate of aging. I think many of the approaches he implements are extreme, but there is real science guiding these decisions. I think many of us could benefit from shifting our lifestyle slightly towards Bryan’s.

Content I Consume - 10/22/2024

Past weeks editions can be found here

 MIT is fully in fall. The leaves are changing and we are getting the last warm days of the year until spring semester. I submitted my first fellowship application. There’s something about checking the first box for PhD applications that feels quite freeing.

Physics

Paper & Machine Learning: Product Manifold Machine Learning Talk

I was invited to speak at the Institute for Artificial Intelligence and Fundamental Interaction’s (IAIFI) thematic discussion of manifold and representation learning. I was sharing my project Product Manifold Machine Learning where we develop methods of representing and processing high energy physics data in curved manifolds (constant curvature Riemannian manifolds).

This approach is motivated by the literature of learned representation and embeddings. This work has been a long effort on my part so I am happy to begin sharing our findings with the research community.

This work was also recently accepted to Conference on Neural Information Processing Systems (NeurIPS) machine learning for the physics science (ML4PS) where I’m first author.

Miscellaneous

Video: Proof of Fermat’s Last Theorem

Pierre de Fermat, first proposed this theorem (shown below) in 1637. He famously wrote that he knew how to prove it, but didn’t have enough space in the margins to write it down. After 357 years, Andrew Wiles finally proved this theorem to be true after a near decade long effort.

This short form documentary outlines Wiles’ largely isolated journey to solving Fermat’s last theorem. I and many others wonder if Fermat’s proof would have been correct, but it’s lost to time.

Content I Consume 10/5/2024

Past weeks editions can be found here

It’s been two months since the last update! As I write this update, I’m sitting on a bench outside MIT’s Stata center. Stata designed by Frank Gehry is a divisive building. When it was first built the roof constantly leaked and MIT famously sued Gehry. If you take a look at the picture I linked above, I think you’d expect the building to have some issues.

The weather in Cambridge is nice this time of the year and I feel a sense of urgency to enjoy the outdoors while I can. My senior year at MIT is underway and I’m approaching my first deadlines for PhD applications. I enjoy the writing, but it’s draining. Adding applications on top of courses and research leaves little room for regular life. Even with this extra work, the semester has been great. So many exciting things going on.  

Physics

Paper: Transforming the Bootstrap: Using Transformers to Compute Scattering Amplitudes in Planar N = 4 Super Yang-Mills Theory

In this work, they teach large ML models to perform manipulation mathematical expressions which arise during quantum field theory calculations. I think this is a great first step towards ML/AI as a tool to perform theory calculations. In terms of the physics, they are calculating something called scattering amplitudes. If we simplify particle collisions, they can be described as,

1. A set of particles come together

2. An interaction between the particles occurs

3. Some set of particles come out of the interaction, not necessarily the same types or numbers of particles that came in

If we fix process (1) and (3), we are really saying that I want to put some fixed set of particles in and get some fixed set of particles out, how likely is this to happen. Scatter amplitudes provide a quantitative description this step (2). The difficulties arise because there are infinitely many ways an interaction could occur. Through quantum field theories, we can determine the relative importance of each of these infinite possibilities.

If we just look at say the top 3 most important ways the interaction could occur we would get a pretty good picture of what happens. However, say we want to test our theory against what experimentalists measure. Sometimes these measurements go up to 10 decimal places. So we need to consider as many possible ways the interaction could occur as a we can. This is where the difficulty arises. There can easily be 10,000+ terms in these calculations. This work develops ML models to aid in manipulating these many term calculations and can hopefully help theorists perform calculation at higher precision.

The training method is interesting. It treats manipulating mathematical expressions as a “translation” task. They find success, but I’m often suspicious when our approach to ML seems overly based on natural language processing and training large language models (like ChatGPT).

This paper is particularly exciting because it was a concerted effort by several prominent researchers in theory (Lance Dixon), experimental physics (Kyle Cranmer), and pure ML (Francois Charton). It seems there’s some real interest in this direction. I’m quite excited by this effort and hope to work on related projects in PhD. I think there’s much work to be done in developing these new tools.

Miscellaneous

Article: MIT class of 2028 to have fewer Black, Latino students after affirmative action ruling

The MIT class of 2028 was the first year of admissions after the SCOTUS ruling outlawing race-based affirmative action. Black and Latino enrolment dropped quite dramatically. I was expecting this drop in minority enrolment, but I was hoping it would come in conjunction with increased enrolment of low-income students. However, this doesn’t seem to be the case.

It’s quite an interesting time to be on campus and see the changing demographics play out in my own communities. These changes are large enough that it’s easy to see even in dorms and walking through campus. Only the coming years will reveal what the MIT community will look like.

Content I Consume - 8/4/2024

I’m finishing up my last month in NYC before heading back to Boston to start my senior year. Now, I’m thinking a lot of how to optimize my time to get results out before apply to grad school as well as how to balance my schedule in the fall. For those interested, I’m applying to grad school hoping to work in ML + some kind of particle phenomenology.

Physics

Chapter: The Renormalization Group

The renormalization group (RG) is a key structure in QFTs and theoretical particle physics. Like many advanced topics, it’s hard to find an introduction that’s self-contained. I’ve read many attempts at explaining the RG before, but I don’t think I truly understood the implications of it’s existence before reading this chapter.

The RG is a mathematical description of how fundamental physics changes as the energy of a system changes (in QFTs). RG highlights beautiful aspects of QFTs as well as some necessary issues with developing a theory of everything.

Miscellaneous

Podcast: The Frankfurt School - Erich Fromm on Love

Eric Fromm is relatively unknown philosopher from the mid 20th century. Fromm is a part of a school of thought called the Frankfurt school. This group was predominately Jewish intellectuals feeling Nazi Germany during the start of WWII. Many of these thinkers contain insightful comments on American culture and it’s impacts both at a society and individual level. One of my favorite works from this group is Dialectic of Enlightenment - Max Horkheimer and Theodor W. Adorno.

Fromm’s book, The Art of Loving, dives into his opinion on love in the western world. Below are some excerpts from the podcast that I particularly enjoy,

On the commodification of love:

… if you’re listening to this, and you’re not currently in love, then barring certain exceptions, for most of us every single attempt you’ve made at love has failed … The answer for Fromm is that most of us are using a horrible strategy. And the slogan for that strategy if there was one is that to find love if you’re not currently in love, become more lovable. Go to the gym. Advance your career. Buy some cool clothes…

Fromm thinks all you’ve really done here is turn yourself into a product to be consumed on what he calls the personality market. And how fitting, to Fromm, that in our modern capitalist societies we’d be so inclined to turn our love lives and the love lives of other people into commodities to be marketed, bought, sold, and traded.

On the fundamental problem with love in western society,

Fromm’s idea says nothing about even considering what the other person can do for me. And this leads him to one of the most important ideas in the entire book The Art of Loving—that you will never be able to love any one person until you can love everyone, because you will always be picking and choosing the people you love in terms of what benefits they can provide to you.

nswood

Content I Consume - 7/20/2024

I moved to NYC for the summer so I’ve been a bit preoccupied and haven’t updated in the past two weeks! In terms of research, I’m pushing to finish two papers this summer, so the pressure is starting to sink in. Progress is slow but steady and I’m starting to understand the realities of being first author.

Machine Learning & AI

Repo: High Powered Computing with SLURM

I primarily work on SLURM based computing clusters. SLURM is great and is a simple system to allocated resources, however, interfacing SLURM with parallelization methods for multi-GPU and multi-node trainings is not so straight forward. In this repo, I combine the methods I’ve pieced together over the past year to deploy Torch DDP and Ray through SLURM. I’m slowly becoming a Ray enthusiast, so more ray content is likely coming in the future!

Physics

Article: World’s Most Accurate and Precise Atomic Clock Pushes New Frontiers in Physics

Researchers at NIST developed an extremely accurate atomic clock. Their clock is so sensitive that it is impacted by the gravitational time dilation due to the presence single human hair (as predicted from Einstein’s theory of general relativity in 1915). The exact creation of this clock uses trapped strontium atoms in an optical lattice. There’s a lot of quantum physics to unpack in this result, but I find it exciting to see how precise our measurements are becoming!

Miscellaneous

Video: How the US is destroying young people’s futures: Scott Galloway

I recently discovered Scott Galloway. He’s a professor at NYU with a uniquely blunt communication style. He presents a number of interesting ideas on reforming US economic structures as well as methods of improving higher education. Some of my favorite ones are:

• Why is social security age based instead of need based? Do we need to be giving the wealthiest group of individuals in the US more money?

• In the same period that Harvard’s endowment increased 4000% and their class size only increased 4%. Who benefits from restricting class sizes at these elite universities?

I don’t endorse all that Galloway says in these, talks but I found his ideas exciting and enjoyed how he seamlessly utilizes data in his presentation.

nswood

Content I Consume - 6/22/2024

I enjoy working on several project at once. Among many of my ML + Physics projects, I’ve started working on my first theoretical physics project. Specifically, I’m working on extending a QCD calculation for to new collider physics process. Getting the chance to explore theoretical work is great before I apply to grad schools and makes the rush to taking QFT 1 my junior year completely worth it.

Machine Learning & AI

Podcast: Mark Zuckerberg - Llama 3, $10B Models, Caesar Augustus, & 1 GW Datacenters - Dwarkesh Podcast

Video: The future of Nuclear = Small, Mobile, Microreactors | Radiant - S3

Dwarkesh’s podcast features many of the big names in AI today. In this episode, he chats with Mark Zuckerberg on a number of topics related to Meta’s AI efforts. To continue AI progress at it’s current rate (without some breakthrough), we must steadily increase compute. This entails relatively large power needs for large data centers.

This thought reminded me of the microreactors of from Radiant. It’s hopeful to see so many new ideas in the nuclear industry. It almost certain that AI datacenters will have dedicated power sources, maybe that power will be nuclear. To my understanding,

Physics

Paper: Generalized Gelation Theory Describes Onset of Online Extremist Support - Manrique et al.

In high school, I was a part of a physics research group. We’d host bi-weekly journal clubs. This paper was one of my favorite articles discussed in those meetings. It models the spread of online extremism through gelation theory. Gelation theory models the properties of gels (like Jell-O). I thought this was an incredibly creative application of physics theory and a wonderful read.

Miscellaneous

Video: Steroids Are Awesome - Jeff Nippard

Jeff Nippard is science-based weightlifting educator. He provides great information supported by peer-review literature. In this video, he covers the impacts of steroid use on several aspects of the human body. I learned they lower your IQ!

Steroid use has skyrocketed in the USA, especially among young people. I’ve been an avid weightlifter for many years and steroids are a constant discussion. I don’t see many easy to access sources of information on the consequences of steroid use. I’m glad this information is getting out there!

Content I Consume - 6/15/2024

This week I returned from a short vacation in Japan and started working full time on research. It’s freeing to have my whole schedule to devote to my work without having to balance my courses.

Machine Learning & AI

Interview: Geoffrey Hinton | On working with Ilya, choosing problems, and the power of intuition

I’ve been enjoying listening to interviews and podcasts as a easy way to keep up with current sentiments of pure ML researchers. In this episode, Geoffrey gives some great insight into the current state of ML research as well as some interesting thoughts on mentorship through his relationship with Ilya.

Physics

Article: The S-Matrix Is the Oracle Physicists Turn To in Times of Crisis - Quanta Magazine

Having taken QFT 1 last semester, it’s exciting to see an examples of how the basic structures in field theories are still used today modern research. Quanta magazine as always provides an amazing overview of the S-matrix suitable for nearly all audiences.

Miscellaneous

Video: Building the World's 1st Hypersonic Airplane | Hermeus - S3

I recently discovered the YouTube channel S^3. He publishes exposes of exciting startups in the USA in both long-form discussions with founders as well as 10-15 min overviews of the companies. This channel is rare as it publishes high quality while remaining utterly undiscovered with only 36k subscribers. One of my favorite episodes covers Hermeus and their quest to develop hypersonic airplanes.