arXiv subjects

Connor Lane

Publications and source records attributed to Connor Lane.

12 recordsLinked to original sources

arXiv2026

Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks

Evaluating large language models (LLMs) for medical applications remains challenging due to benchmark saturation, limited data accessibility, and insufficient coverage of relevant tasks. Existing suites have either saturated, heavily depend on restricted datasets, or lack comprehensive model coverage. We introduce Medmarks, a fully open-source evaluation suite with 30 benchmarks spanning question answering, information extraction, medical calculations, and open-ended clinical reasoning. We perform a systematic evaluation of 61 models across 71 configurations using verifiable metrics and LLM-as-a-Judge. Our results show that frontier reasoning models (Gemini 3 Pro Preview, GPT-5.1, & GPT-5.2) achieve the highest performance across both benchmarks, most frontier proprietary models are significantly more token efficient than open-weight alternatives, medically fine-tuned models outperform their generalist counterparts, and that models are susceptible to answer-order bias (particularly smaller models and Grok 4). A subset of our evals (Medmarks-T) can be directly used as reinforcement learning environments to post-train LLMs for medical reasoning. Code is available at https://github.com/MedARC-AI/Medmarks

cs.CL↗

arXiv2025

Scaling Vision Transformers for Functional MRI with Flat Maps

We study the problem of training self-supervised foundation models for functional MRI. Our main contributions are: (1) we introduce a new model family (CortexMAE) trained using the masked autoencoder framework on 2.1K hours of open fMRI data, and (2) we release the first open evaluation suite (Brainmarks) for fMRI foundation models. Our core innovation is simple: we adapt the Vision Transformer to fMRI by first converting each 3D fMRI volume to a 2D map using a cortical flat map projection. We directly compare flat maps to both parcellation and volume-based representations. While each has its advantages, flat maps generally perform best. We perform the first systematic scaling analysis for fMRI and observe strict power law scaling, albeit with limits. Finally, we use Brainmarks to do controlled benchmark comparisons. On subject-level trait prediction, we report a challenging null result: no single model achieves clear state-of-the-art performance. Moreover, all models struggle to outperform a simple functional connectivity baseline. On cognitive state decoding, we observe more robust performance, and in this setting our CortexMAE family outperforms prior models by a large margin. Code, models, and datasets are available at https://github.com/MedARC-AI/CortexMAE and https://github.com/MedARC-AI/Brainmarks.

cs.CV↗

arXiv2025

Predicting Brain Responses To Natural Movies With Multimodal LLMs

We present MedARC's team solution to the Algonauts 2025 challenge. Our pipeline leveraged rich multimodal representations from various state-of-the-art pretrained models across video (V-JEPA2), speech (Whisper), text (Llama 3.2), vision-text (InternVL3), and vision-text-audio (Qwen2.5-Omni). These features extracted from the models were linearly projected to a latent space, temporally aligned to the fMRI time series, and finally mapped to cortical parcels through a lightweight encoder comprising a shared group head plus subject-specific residual heads. We trained hundreds of model variants across hyperparameter settings, validated them on held-out movies and assembled ensembles targeted to each parcel in each subject. Our final submission achieved a mean Pearson's correlation of 0.2085 on the test split of withheld out-of-distribution movies, placing our team in fourth place for the competition. We further discuss a last-minute optimization that would have raised us to second place. Our results highlight how combining features from models trained in different modalities, using a simple architecture consisting of shared-subject and single-subject components, and conducting comprehensive model selection and ensembling improves generalization of encoding models to novel movie stimuli. All code is available on GitHub.

cs.CV↗

arXiv2025

Dispersive estimates for Dirac Operators in dimension four with obstructions at threshold energies

We investigate $L^1\to L^\infty$ dispersive estimates for the Dirac equation with a potential in four spatial dimensions. We classify the structure of the obstructions at the thresholds as being composed of an at most two dimensional space of resonances per threshold, and finitely many eigenfunctions. Similar to the Schr\"odinger evolution, we prove the natural $t^{-2}$ decay rate when the thresholds are regular. When there is a threshold resonance or eigenvalue, we show that there is a time dependent, finite rank operator satisfying $\|F_t\|_{L^1\to L^\infty}\lesssim (\log t)^{-1}$ for $t>2$ such that $$ \|e^{it\mathcal H}P(\mathcal H)-F_t\|_{L^1\to L^\infty}\lesssim t^{-1} \quad \text{for } t>2, $$ with $P$ a projection onto a subspace of the absolutely continuous spectrum in a small neighborhood of the thresholds. We further show that the operator $F_t=0$ if there is a threshold eigenvalue but no threshold resonance. We pair this with high energy bounds for the evolution and provide a complete description of the dispersive bounds.

math.AP↗

arXiv2024

AGM aquariums and elliptic curves over arbitrary finite fields

In this paper, we define a version of the arithmetic-geometric mean (AGM) function for arbitrary finite fields $\mathbb{F}_q$, and study the resulting AGM graph with points $(a,b) \in \mathbb{F}_q \times \mathbb{F}_q$ and directed edges between points $(a,b)$, $(\frac{a+b}{2},\sqrt{ab})$ and $(a,b)$, $(\frac{a+b}{2},-\sqrt{ab})$. The points in this graph are naturally associated to elliptic curves over $\mathbb{F}_q$ in Legendre normal form, with the AGM function defining a 2-isogeny between the associated curves. We use this correspondence to prove several results on the structure, size, and multiplicity of the connected components in the AGM graph.

math.NT↗

arXiv2024

Subspaces spanned by eigenforms with nonvanishing twisted central $L$-values

In this paper, we construct explicit spanning sets for two spaces. One is the subspace generated by integral-weight Hecke eigenforms with nonvanishing quadratic twisted central $L$-values. The other is a subspace generated by half-integral weight Hecke eigenforms with certain nonvanishing Fourier coefficients. Along the way, we show that these subspaces are isomorphic via the Shimura lift.

math.NT↗

arXiv2024

The Massless Dirac Equation in Three Dimensions: Dispersive estimates and zero energy obstructions

We investigate dispersive estimates for the massless three dimensional Dirac equation with a potential. In particular, we show that the Dirac evolution satisfies a $\langle t\rangle^{-1}$ decay rate as an operator from $L^1$ to $L^\infty$ regardless of the existence of zero energy eigenfunctions. We also show this decay rate may be improved to $\langle t\rangle ^{-1-\gamma}$ for any $0\leq \gamma < 1/2$ at the cost of spatial weights. This estimate, along with the $L^2$ conservation law allows one to deduce a family of Strichartz estimates in the case of a threshold eigenvalue. We classify the structure of threshold obstructions as being composed of zero energy eigenfunctions. Finally, we show the Dirac evolution is bounded for all time with minimal requirements on the decay of the potential and smoothness of initial data.

math.AP↗

arXiv2023

Asymptotic Distribution of Residues in Pascal's Triangle mod $p$

Fix a prime $p$ and define $T_p(n)$ to be the number of nonzero residues in the $n$th row of pascal's triangle mod $p$, and define $\phi_p(n)$ to be the number of nonzero residues in the first $n$ rows of pascal's triangle mod $p$. We generalize these to sequences $T_\chi(n)$ and $\phi_\chi(n)$ for a Dirichlet character $\chi$ of modulus $p$. We prove many properties of these sequences that generalize those of $T_p(n)$ and $\phi_p(n)$. Define $A_n(r)$ to be the number of occurrences of $r$ in the first $n$ rows of Pascal's triangle mod $p$. Guy Barat and Peter Grabner showed that for all primes $p$ and nonzero residues $r$, $A_n(r)\sim \frac{1}{p-1}\phi_p(n)$. We provide an alternative proof of this fact that yields explicit bounds on the error term. We also discuss the distribution of $A_p(r)$.

math.NT↗

arXiv2023

A Parameter-efficient Multi-subject Model for Predicting fMRI Activity

This is the Algonauts 2023 submission report for team "BlobGPT". Our model consists of a multi-subject linear encoding head attached to a pretrained trunk model. The multi-subject head consists of three components: (1) a shared multi-layer feature projection, (2) shared plus subject-specific low-dimension linear transformations, and (3) a shared PCA fMRI embedding. In this report, we explain these components in more detail and present some experimental results. Our code is available at https://github.com/cmi-dair/algonauts23.

cs.CV↗

arXiv2019

On the Regularization Properties of Structured Dropout

Dropout and its extensions (eg. DropBlock and DropConnect) are popular heuristics for training neural networks, which have been shown to improve generalization performance in practice. However, a theoretical understanding of their optimization and regularization properties remains elusive. Recent work shows that in the case of single hidden-layer linear networks, Dropout is a stochastic gradient descent method for minimizing a regularized loss, and that the regularizer induces solutions that are low-rank and balanced. In this work we show that for single hidden-layer linear networks, DropBlock induces spectral k-support norm regularization, and promotes solutions that are low-rank and have factors with equal norm. We also show that the global minimizer for DropBlock can be computed in closed form, and that DropConnect is equivalent to Dropout. We then show that some of these results can be extended to a general class of Dropout-strategies, and, with some assumptions, to deep non-linear networks when Dropout is applied to the last layer. We verify our theoretical claims and assumptions experimentally with commonly used network architectures.

cs.LG↗

arXiv2017

Dropout as a Low-Rank Regularizer for Matrix Factorization

Regularization for matrix factorization (MF) and approximation problems has been carried out in many different ways. Due to its popularity in deep learning, dropout has been applied also for this class of problems. Despite its solid empirical performance, the theoretical properties of dropout as a regularizer remain quite elusive for this class of problems. In this paper, we present a theoretical analysis of dropout for MF, where Bernoulli random variables are used to drop columns of the factors. We demonstrate the equivalence between dropout and a fully deterministic model for MF in which the factors are regularized by the sum of the product of squared Euclidean norms of the columns. Additionally, we inspect the case of a variable sized factorization and we prove that dropout achieves the global minimum of a convex approximation problem with (squared) nuclear norm regularization. As a result, we conclude that dropout can be used as a low-rank regularizer with data dependent singular-value thresholding.

cs.LG↗

arXiv2017

An Analysis of Dropout for Matrix Factorization

Dropout is a simple yet effective algorithm for regularizing neural networks by randomly dropping out units through Bernoulli multiplicative noise, and for some restricted problem classes, such as linear or logistic regression, several theoretical studies have demonstrated the equivalence between dropout and a fully deterministic optimization problem with data-dependent Tikhonov regularization. This work presents a theoretical analysis of dropout for matrix factorization, where Bernoulli random variables are used to drop a factor, thereby attempting to control the size of the factorization. While recent work has demonstrated the empirical effectiveness of dropout for matrix factorization, a theoretical understanding of the regularization properties of dropout in this context remains elusive. This work demonstrates the equivalence between dropout and a fully deterministic model for matrix factorization in which the factors are regularized by the sum of the product of the norms of the columns. While the resulting regularizer is closely related to a variational form of the nuclear norm, suggesting that dropout may limit the size of the factorization, we show that it is possible to trivially lower the objective value by doubling the size of the factorization. We show that this problem is caused by the use of a fixed dropout rate, which motivates the use of a rate that increases with the size of the factorization. Synthetic experiments validate our theoretical findings.

cs.LG↗