arXiv subjects

Justin Chen

Publications and source records attributed to Justin Chen.

At least 19 recordsLinked to original sources

arXiv2026

Accelerating Scientific Research with Gemini in the Real-World

We present an extension and comprehensive real-world validation of Co-Scientist, a Gemini-based multi-agent system designed to accelerate end-to-end scientific research across hypothesis generation, experimentation, and manuscript generation. Moving beyond in silico hypothesis generation, this specialized configuration transitions Co-Scientist into an execution-grounded research partner advancing closed-loop scientific workflows across materials science, biology, and computer science. In materials science, Co-Scientist interfaced with a semi-automated chemical vapor deposition reactor to design a safe precursor route for MXenes; experimental execution produced a lamellar 2D material sharing key structural similarities with the Ti3C2Tx MXene lattice, although further experiments are needed to confirm the atomic structure. Leveraging Gemini 3 Deep Think for rapid, lab-in-the-loop execution, it also tailored growth recipes to laboratory constraints in minutes, enabling single-attempt growth of monolayer MoS2, MoSe2, and WS2 semiconductors. In biology, Co-Scientist predicted emergent swarming phenotypes of engineered E. coli across inducer (IPTG) gradients from sparse imaging data, quantitatively matching unpublished wet-lab morphological measurements. In computer science, Co-Scientist autonomously discovered an inference-time scaling architecture that outperformed six frontier models on HealthBench (Hard and Professional) while reducing potential clinical harm under blinded physician evaluation. Finally, a double-blind study of end-to-end generated papers with 30 domain experts across 450 reviews demonstrates that Co-Scientist's reliability modules reduce hallucination and plagiarism while improving research safety. Together, these results demonstrate progress toward closed-loop multi-agent scientific AI systems capable of accelerating real-world scientific discovery.

cs.AI↗

arXiv2026

ResidencyRL: Reinforcement Learning in Simulated Clinical Environments

In medical education, physicians convert academic knowledge into clinical expertise through residency: years of training across thousands of encounters, with diverse sources of feedback and progressively greater autonomy. Much of clinical reasoning relies on the patient encounter, a dialogue in which a clinician elicits history, refines diagnostic hypotheses, and decides management under uncertainty. While large language models (LLMs) excel on static medical benchmarks, methods to optimize the full sequence of clinical decisions remain underdeveloped. We present ResidencyRL, a reinforcement learning (RL) method for training clinical artificial intelligence (AI) agents through simulated multi-turn clinical encounters (up to 60 dialogue turns and 8 tool calls per trajectory). ResidencyRL pairs the policy agent with LLM simulators capable of complex, adversarial behaviors, training against a structured reward aligned to diagnostic accuracy, management quality, communication, documentation, and safety. On held-out evaluations, the ResidencyRL agent improves diagnostic accuracy by 7.0% under adversarial conditions (88.0% vs. 81.0%) and reduces missed red flag rates by 31%, demonstrating rigorous mitigation of premature closure. Blinded expert clinicians validated these gains, preferring the trained agent in 87.6% of side-by-side comparisons. The procedural competencies transfer to unseen benchmarks: the agent outperforms the base model across all six clinical axes of the AMIE multi-visit benchmark, and shows consistent directional improvements on AgentClinic and CRAFT-MD. Our findings demonstrate that sequential clinical decision-making can be effectively learned through multi-turn RL in simulation, yielding robust, generalizable capabilities, paving the way towards clinical mastery. Prospective validation with real-world workflows remains necessary to establish clinical utility.

cs.AI↗

arXiv2025

Degeneracy and Sato-Tate groups of $y^2=x^{p^2}-1$

We say that an abelian variety is degenerate if its Hodge ring is not generated by divisor classes. Degeneracy leads to some interesting challenges when computing Sato-Tate groups, and there are currently few examples and techniques presented in the literature. In this paper we focus on the Jacobians of the family of curves $C_{p^2}: y^2=x^{p^2}-1$, where $p$ is an odd prime. Using a construction developed by Shioda in the 1980s, we are able to characterize so-called indecomposable Hodge classes as well as the Sato-Tate groups of these Jacobian varieties. Our work is inspired by computation, and examples and methods are described throughout the paper.

math.NT↗

arXiv2025

MedGemma Technical Report

Artificial intelligence (AI) has significant potential in healthcare applications, but its training and deployment faces challenges due to healthcare's diverse data, complex tasks, and the need to preserve privacy. Foundation models that perform well on medical tasks and require less task-specific tuning data are critical to accelerate the development of healthcare AI applications. We introduce MedGemma, a collection of medical vision-language foundation models based on Gemma 3 4B and 27B. MedGemma demonstrates advanced medical understanding and reasoning on images and text, significantly exceeding the performance of similar-sized generative models and approaching the performance of task-specific models, while maintaining the general capabilities of the Gemma 3 base models. For out-of-distribution tasks, MedGemma achieves 2.6-10% improvement on medical multimodal question answering, 15.5-18.1% improvement on chest X-ray finding classification, and 10.8% improvement on agentic evaluations compared to the base models. Fine-tuning MedGemma further improves performance in subdomains, reducing errors in electronic health record information retrieval by 50% and reaching comparable performance to existing specialized state-of-the-art methods for pneumothorax classification and histopathology patch classification. We additionally introduce MedSigLIP, a medically-tuned vision encoder derived from SigLIP. MedSigLIP powers the visual understanding capabilities of MedGemma and as an encoder achieves comparable or better performance than specialized medical image encoders. Taken together, the MedGemma collection provides a strong foundation of medical image and text capabilities, with potential to significantly accelerate medical research and development of downstream applications. The MedGemma collection, including tutorials and model weights, can be found at https://goo.gle/medgemma.

cs.AI↗

arXiv2025

Fringe around a Beet Slice: Wetting-induced Dimple in a Thin Liquid Film

When a slice of beet is placed on a plate with a thin layer of beet juice, one can observe a clear fringe around the beet, where the color is more translucent than the rest of the juice. The hypotheses in literature were inconsistent and limited, which motivated us to revisit this phenomenon. Using a motorized confocal displacement sensor, we measured the temporal evolution of the liquid surface profile across the fringe. Our findings suggest that a suction flow, induced by the capillary rise of the contact line, causes a dimple - a small concave depression - to form on the liquid surface. While surface tension and gravity tends to smooth out the dimple, viscous drag acts against them if the liquid film is sufficiently thin. Our scaling analysis correctly estimates the dependence of dimple lifetime on liquid properties and film thickness. We also capture the dimple formation dynamics by numerically solving the lubrication equation with the Young-Laplace equation. This work provides a new interpretation for a common phenomenon.

cond-mat.soft↗

arXiv2024

Adversarial training of Keyword Spotting to Minimize TTS Data Overfitting

The keyword spotting (KWS) problem requires large amounts of real speech training data to achieve high accuracy across diverse populations. Utilizing large amounts of text-to-speech (TTS) synthesized data can reduce the cost and time associated with KWS development. However, TTS data may contain artifacts not present in real speech, which the KWS model can exploit (overfit), leading to degraded accuracy on real speech. To address this issue, we propose applying an adversarial training method to prevent the KWS model from learning TTS-specific features when trained on large amounts of TTS data. Experimental results demonstrate that KWS model accuracy on real speech data can be improved by up to 12% when adversarial loss is used in addition to the original KWS loss. Surprisingly, we also observed that the adversarial setup improves accuracy by up to 8%, even when trained solely on TTS and real negative speech data, without any real positive examples.

cs.SD↗

arXiv2024

Utilizing TTS Synthesized Data for Efficient Development of Keyword Spotting Model

This paper explores the use of TTS synthesized training data for KWS (keyword spotting) task while minimizing development cost and time. Keyword spotting models require a huge amount of training data to be accurate, and obtaining such training data can be costly. In the current state of the art, TTS models can generate large amounts of natural-sounding data, which can help reducing cost and time for KWS model development. Still, TTS generated data can be lacking diversity compared to real data. To pursue maximizing KWS model accuracy under the constraint of limited resources and current TTS capability, we explored various strategies to mix TTS data and real human speech data, with a focus on minimizing real data use and maximizing diversity of TTS output. Our experimental results indicate that relatively small amounts of real audio data with speaker diversity (100 speakers, 2k utterances) and large amounts of TTS synthesized data can achieve reasonably high accuracy (within 3x error rate of baseline), compared to the baseline (trained with 3.8M real positive utterances).

cs.SD↗

arXiv2023

Local dual spaces and primary decomposition

Generalizing the concept of the Macaulay inverse system, we introduce a way to describe localizations of an ideal in a polynomial ring. This leads to an approach to the differential primary decomposition as a description of the affine scheme defined by the ideal.

math.AC↗

arXiv2023

VisGrader: Automatic Grading of D3 Visualizations

Manually grading D3 data visualizations is a challenging endeavor, and is especially difficult for large classes with hundreds of students. Grading an interactive visualization requires a combination of interactive, quantitative, and qualitative evaluation that are conventionally done manually and are difficult to scale up as the visualization complexity, data size, and number of students increase. We present VisGrader, a first-of-its kind automatic grading method for D3 visualizations that scalably and precisely evaluates the data bindings, visual encodings, interactions, and design specifications used in a visualization. Our method enhances students learning experience, enabling them to submit their code frequently and receive rapid feedback to better inform iteration and improvement to their code and visualization design. We have successfully deployed our method and auto-graded D3 submissions from more than 4000 students in a visualization course at Georgia Tech, and received positive feedback for expanding its adoption.

cs.HC↗

arXiv2023

Representing matroids via pasture morphisms

Using the framework of pastures and foundations of matroids developed by Baker-Lorscheid, we give algorithms to: (i) compute the foundation of a matroid, and (ii) compute all morphisms between two pastures. Together, these provide an efficient method of solving many questions of interest in matroid representations, including orientability, non-representability, and computing all representations of a matroid over a finite field.

math.CO↗

arXiv2023

Units of hyperelliptic curves over $\mathbb{F}_2$

We study unit groups of rings of the form $\mathbb{F}_2[x,y]/(y^2 + gy + h)$, for $g, h \in \mathbb{F}_2[x]$ -- in particular, the question of (non)triviality of such unit groups. Up to automorphisms of $\mathbb{F}_2[x,y]$ we classify such rings into 3 distinct types. For 2 of the types we show that the unit group is always trivial, and conjecture that the unit group is always nontrivial for the 3rd type. We provide support for this conjecture both theoretically and computationally, via an algorithm that has been used to compute units in large degrees.

math.AC↗

arXiv2022

On the ideal avoidance property

In this article, we investigate the avoidance property of ideals and rings. Among the main results, a general version of the avoidance lemma is formulated. It is shown that every idempotent ideal (and hence every pure ideal) has avoidance. The avoidance property of arbitrary direct products of avoidance rings is characterized. It is shown that every overring of an avoidance domain is an avoidance domain. Next, we show that every avoidance $\mathbb{N}$-graded ring whose base subring is a finite field is a PIR. It is also proved that the avoidance property is preserved under flat ring epimorphisms. Dually, we formulate a notion of strong avoidance, and show that it is reflected by pure morphisms.

math.AC↗

arXiv2021

Towards Automatic Grading of D3.js Visualizations

Manually grading D3 data visualizations is a challenging endeavor, and is especially difficult for large classes with hundreds of students. Grading an interactive visualization requires a combination of interactive, quantitative, and qualitative evaluation that are conventionally done manually and are difficult to scale up as the visualization complexity, data size, and number of students increase. We present a first-of-its kind automatic grading method for D3 visualizations that scalably and precisely evaluates the data bindings, visual encodings, interactions, and design specifications used in a visualization. Our method has shown potential to enhance students' learning experience, enabling them to submit their code frequently and receive rapid feedback to better inform iteration and improvement to their code and visualization design. Our method promotes consistent grading and enables instructors to dedicate more focus to assist students in gaining visualization knowledge and experience. We have successfully deployed our method and auto-graded D3 submissions from more than 1000 undergraduate and graduate students in Georgia Tech's CSE6242 Data and Visual Analytics course, and received positive feedback and encouragement for expanding its adoption.

cs.HC↗

arXiv2021

Sums of squares, Hankel index, and almost real rank

The Hankel index of a real variety $X$ is an invariant that quantifies the difference between nonnegative quadrics and sums of squares on $X$. In [5], the authors proved an intriguing bound on the Hankel index in terms of the Green-Lazarsfeld index, which measures the "linearity" of the minimal free resolution of the ideal of $X$. In all previously known cases this bound was tight. We provide the first class of examples where the bound is not tight; in fact the difference between Hankel index and Green-Lazarsfeld index can be arbitrarily large. Our examples are outer projections of rational normal curves, where we identify the center of projection with a binary form $F$. The Green-Lazarsfeld index of the projected curve is given by the complex Waring border rank of $F$ [15]. We show that the Hankel index is given by the "almost real" rank of $F$, which is a new notion that comes from decomposing $F$ as a sum of powers of almost real forms. We determine the range of possible and typical almost real ranks for binary forms.

math.AG↗

arXiv2021

Primary decomposition of modules: a computational differential approach

We study primary submodules and primary decompositions from a differential and computational point of view. Our main theoretical contribution is a general structure theory and a representation theorem for primary submodules of an arbitrary finitely generated module over a polynomial ring. We characterize primary submodules in terms of differential operators and punctual Quot schemes. Moreover, we introduce and implement an algorithm that computes a minimal differential primary decomposition for a module.

math.AC↗

arXiv2021

Computing multiplicity sequences

The MultiplicitySequence package for Macaulay2 computes the multiplicity sequence of a graded ideal in a standard graded ring over a field, as well as several invariants of monomial ideals related to integral dependence. We discuss two strategies implemented for computing multiplicity sequences: one via the bivariate Hilbert polynomial, and the other via the technique of general elements.

math.AC↗

arXiv2021

Noetherian Operators in Macaulay2

A primary ideal in a polynomial ring can be described by the variety it defines and a finite set of Noetherian operators, which are differential operators with polynomial coefficients. We implement both symbolic and numerical algorithms to produce such a description in various scenarios as well as routines for studying affine schemes through the prism of Noetherian operators and Macaulay dual spaces.

math.AC↗

arXiv2020

FrUITeR: A Framework for Evaluating UI Test Reuse

UI testing is tedious and time-consuming due to the manual effort required. Recent research has explored opportunities for reusing existing UI tests from an app to automatically generate new tests for other apps. However, the evaluation of such techniques currently remains manual, unscalable, and unreproducible, which can waste effort and impede progress in this emerging area. We introduce FrUITeR, a framework that automatically evaluates UI test reuse in a reproducible way. We apply FrUITeR to existing test-reuse techniques on a uniform benchmark we established, resulting in 11,917 test reuse cases from 20 apps. We report several key findings aimed at improving UI test reuse that are missed by existing work.

cs.SE↗