Bronx Science '27
Senior at The Bronx High School of Science. When I'm not in class or in the lab, you'll find me playing pickup soccer with my friends, grinding Rocket League, a game where cars play soccer, or researching how Napoleon brought Europe to its knees.
01
I grew up between France and the United States, which shaped the way I move through most things: switching languages mid-sentence, adapting quickly, and being able to connect with people.
Most of the research I have done traces back to when Dr. Yan Yu took a high school student under her wing at St. John's University. She taught me what her data actually meant and why it mattered, how to work with it, and how to turn it into something with an impact. Through her lessons I have been able to apply the same principles to research well outside her field of neuroscience.
Outside of that, I play forward in MLS Next for Metropolitan Oval Academy and captain the varsity team at Bronx Science, where I was top scorer last season. I am co-captain of the varsity tennis team and a longtime violinist and pianist.
02
Two diplomas in parallel, and a course load built around mathematics, physics, and computation.
Four-year selective research track. Senior coursework includes Ordinary Differential Equations, AP Physics C: Mechanics, AP Statistics, AP Macroeconomics and Microeconomics, and AP English Literature.
Complete French national curriculum in literature, history, and mathematics, carried in tandem with the U.S. diploma across all four years of high school.
AP Scholar with Distinction. Five additional exams scheduled for May 2027.
Supervised machine learning, CNNs, and data pipelines with Stanford and MIT researchers, followed by one-on-one research mentorship on EEG machine learning analysis.
Quantum computing course covering qubits, gates, entanglement, and quantum algorithms.
Financial analysis, markets, and corporate valuation.
03
Computational neuroscience, quantum information, and quantitative methods.
November 2024 to Present
Lead project on machine learning analysis of musical pitch and rhythm processing in bilingual and monolingual children. I analyze resting-state and speech EEG, handling artifact rejection, filtering, and baseline correction, then extract ERPs and oscillatory features. I map neural coherence and network development using BESA and wavelet analysis, and train supervised models in scikit-learn to predict detection accuracy directly from EEG.
May 2026
Music and language experience both shape auditory development, but it remains unclear how language background specifically affects music processing. Children aged five to ten from Mandarin, Spanish, French, Standard American English, and African American English speaking households completed a passive listening oddball task covering six kinds of musical change: intensity, pitch, rhythm, timbre, slide, and location. Event-related potentials were recorded from 65 scalp sites and connectivity was analyzed across frequency bands from 1 to 40 Hz.
I then used those functional connectivity measures as input features for classification models trained to distinguish children by language background, comparing support vector machines, random forests, and neural networks across acoustic features. The models captured real differences in connectivity patterns, which suggests language background leaves a measurable mark on how music processing networks develop.
Summer 2026
Selected for an intensive four-week research course applying mathematical models and data analysis to financial markets and quantitative trading strategies, covering R and RStudio alongside Bloomberg Market Concepts and Capital IQ Pro certifications. I carried a group assignment from the course forward into a paper of my own, asking whether Tesla's well known volatility actually pays investors for the risk it imposes.
Working in R with roughly 2,190 trading days of log returns for TSLA, IBM, and SPY from 2018 onward, I found Tesla's return variance runs about four times IBM's, which an F-test confirms, and a GARCH(1,1) model shows that volatility is not only higher but less predictable period to period. The surprise came in the regressions. IBM on its own is a significant positive predictor of Tesla's daily return, but once SPY enters the model the coefficient flips negative and stays significant, so the two are not linked to each other at all; they simply react to the same market in opposite directions.
None of Tesla's extra risk looks well compensated. Its Sharpe ratio falls below IBM's in many quarters, its worst peak-to-trough drawdown reaches 80 percent against IBM's 43, and its beta climbs from 1.33 to 2.26 across the sample rather than settling as the company matures.
September 2025
Mandarin uses pitch contour to carry word meaning, so tone perception puts unusual demands on neural timing. I asked how those networks mature in bilingual Mandarin-English speakers, comparing participants across three age groups: children aged 5 to 7, children aged 8 to 10, and adults. Each completed a passive lexical tone listening task and a resting-state recording on a 65-channel EEG system sampled at 1000 Hz, analyzed with coherence and phase-locking value.
The trajectory turned out to be non-linear. Older children showed the strongest and most widespread connectivity, stronger than adults, whose sparser networks likely reflect greater automaticity in speech processing rather than immaturity. Resting-state coherence barely changed from early childhood to adulthood, which means task-based EEG is the more sensitive measure of developmental change. All data were collected under an approved IRB protocol.
June to July 2025
An entanglement witness can tell an entangled state from a separable one without reconstructing the full density matrix, which makes it appealing for real hardware. We asked how much noise it takes to break that. In Cirq we prepared a Φ+ Bell state, measured correlations along the X, Y, and Z axes, and evaluated the witness as one minus the sum of those three expectation values, where a negative result confirms entanglement. Every trial ran 1,000 shots, repeated three times at each depolarizing noise level from 5 to 25 percent.
The witness held at 5 percent noise and barely survived at 15 percent, then failed at 25 percent by calling a genuinely entangled state separable. That puts the failure threshold between 15 and 25 percent. One pattern was worth the whole project: the Z component stayed consistently stronger than X and Y, because a Φ+ state is already aligned with the Z axis and needs no rotation gate to measure. A single extra gate per qubit was enough to measurably degrade the result, which is a concrete argument against relying on witnesses in NISQ devices.
In the same program I also simulated Grover's search algorithm combined with BB84 quantum key distribution, running those circuits on IBM quantum hardware and analyzing decoherence and measurement noise.
04
Work I have built, published, or presented.
We built a CNN in TensorFlow to classify 2,700 aerial and ground-level images as fire or no-fire, then tested how dataset size, training epochs, and image resolution each affect detection accuracy. Accuracy climbed steadily with dataset size, from chance level on 10 images to roughly 75 percent on 700, with the best runs classifying the test set perfectly from epoch 15 onward.
The result we did not expect: 100x100 pixel images outperformed both 10x10 and 250x250. Higher resolution is usually assumed to help, but past a point the added detail introduced noise rather than signal. The project took 2nd place at the Terra NYC STEM Fair, an ISEF-affiliated regional championship, and was submitted to the Conrad Challenge.
Ongoing Regeneron STS project asking whether language background shapes how children process music, using connectivity features from EEG as inputs to classification models. Presented at the Acoustical Society of America in 2026.
View the abstractSimulated Bell states in Cirq across noise levels from 5 to 25 percent to find where entanglement witnesses stop working, and traced the asymmetry between measurement axes back to a single extra rotation gate.
Read the paperEight years of daily returns in R, tested with an F-test, GARCH(1,1), CAPM, and mean-variance portfolio theory. Tesla carries four times IBM's variance without the risk-adjusted return to justify it, and its apparent link to IBM reverses once the market is controlled for.
Read the paper
Founded and lead a club of more than 20 members. I run weekly lectures on superposition, entanglement, and where quantum mechanics collides with relativity, design Python and Qiskit coding workshops, and coordinate guest speakers.
Tutored Bronx middle schoolers preparing for the SHSAT through the Science for Bronx Club, roughly four hours a week across two years. Three of us were often responsible for twenty-five to thirty students at once, and one-on-one help did not scale: sessions ran chaotic and most questions went unanswered.
I pushed the club toward small rotating groups instead, each session built around a single math or reading skill and planned in advance rather than improvised. Students engaged more, could track their own progress, and needed less hand-holding as they worked through practice problems. The model held as membership grew.
As a Bronx Science Senator in grades 10 and 12, I built a shared resource guide for my classmates: one place to compare the school's clubs and activities, and a practical walkthrough for finding research internships with university professors.
I started Model UN on the novice team and competed on varsity from grades 10 through 12. At StuyMUN I represented France in a committee debating the costs of colonialism, a difficult seat for a former colonial power. Instead of defending France's position outright, I worked toward a middle ground that both former colonizers and formerly colonized nations could accept, and was named Honorable Delegate. I also helped organize SciMUN competitions.
05
Soccer, tennis, and twenty-two years of instruments between two hands.
Highlight reel. Forward and attacking midfielder, MLS Next and Bronx Science varsity.
Antonio Vivaldi, Violin Concerto in A minor, Op. 3 No. 6, RV 356, first movement. Harlem School of the Arts.
06
The Impact of Image Classification Using CNN on Wildfire Detection
190th International Meeting, Philadelphia
National Honor Society member
International conference at Gallaudet University
Represented France in a committee on the costs of colonialism
ISEF-affiliated regional championship at NYU Tandon
07
I am always glad to talk about research, coursework, or anything at the edge of math and computing. The fastest way to reach me is email.
elliotblin2509@gmail.com