Tootfinder

Private Continual Counting of Unbounded Streams
Ben Jacobsen, Kassem Fawaz
https://arxiv.org/abs/2506.15018 https://arxiv.org/pdf/250…

Private Continual Counting of Unbounded Streams
We study the problem of differentially private continual counting in the unbounded setting where the input size $n$ is not known in advance. Current state-of-the-art algorithms based on optimal instantiations of the matrix mechanism cannot be directly applied here because their privacy guarantees only hold when key parameters are tuned to $n$. Using the common `doubling trick' avoids knowledge of $n$ but leads to suboptimal and non-smooth error. We solve this problem by introducing novel matrix…

Heisenberg limited multiple eigenvalue estimation via off-the-grid compressed sensing
Davide Castaldo, Stefano Corni
https://arxiv.org/abs/2507.12438 https…

Heisenberg limited multiple eigenvalue estimation via off-the-grid compressed sensing
Quantum phase estimation is the flagship algorithm for quantum simulation on fault-tolerant quantum computers. We demonstrate that an \emph{off-grid} compressed sensing protocol, combined with a state-of-the-art signal classification method, enables the simultaneous estimation of multiple eigenvalues of a unitary matrix using the Hadamard test while sampling only a few percent of the full autocorrelation function. Our numerical evidence indicates that the proposed algorithm achieves the Heisenb…

Eigenstate Thermalization Hypothesis (ETH) for off-diagonal matrix elements in integrable spin chains
Federico Rottoli, Vincenzo Alba
https://arxiv.org/abs/2505.23602

Eigenstate Thermalization Hypothesis (ETH) for off-diagonal matrix elements in integrable spin chains
We investigate off-diagonal matrix elements of local operators in integrable spin chains, focusing on the isotropic spin-$1/2$ Heisenberg chain ($XXX$ chain). We employ state-of-the-art Algebraic Bethe Ansatz results, which allow us to efficiently compute matrix elements of operators with support up to two sites between generic energy eigenstates. We consider both matrix elements between eigenstates that are in the same thermodynamic macrostate, as well as eigenstates that belong to different m…

Low-rank Momentum Factorization for Memory Efficient Training
Pouria Mahdavinia, Mehrdad Mahdavi
https://arxiv.org/abs/2507.08091 https://arxiv.org/pdf/2507.08091 https://arxiv.org/html/2507.08091
arXiv:2507.08091v1 Announce Type: new
Abstract: Fine-tuning large foundation models presents significant memory challenges due to stateful optimizers like AdamW, often requiring several times more GPU memory than inference. While memory-efficient methods like parameter-efficient fine-tuning (e.g., LoRA) and optimizer state compression exist, recent approaches like GaLore bridge these by using low-rank gradient projections and subspace moment accumulation. However, such methods may struggle with fixed subspaces or computationally costly offline resampling (e.g., requiring full-matrix SVDs). We propose Momentum Factorized SGD (MoFaSGD), which maintains a dynamically updated low-rank SVD representation of the first-order momentum, closely approximating its full-rank counterpart throughout training. This factorization enables a memory-efficient fine-tuning method that adaptively updates the optimization subspace at each iteration. Crucially, MoFaSGD leverages the computed low-rank momentum factors to perform efficient spectrally normalized updates, offering an alternative to subspace moment accumulation. We establish theoretical convergence guarantees for MoFaSGD, proving it achieves an optimal rate for non-convex stochastic optimization under standard assumptions. Empirically, we demonstrate MoFaSGD's effectiveness on large language model alignment benchmarks, achieving a competitive trade-off between memory reduction (comparable to LoRA) and performance compared to state-of-the-art low-rank optimization methods. Our implementation is available at https://github.com/pmahdavi/MoFaSGD.
toXiv_bot_toot

This https://arxiv.org/abs/2506.03074 has been replaced.
initial toot: https://mastoxiv.page/@arXiv_sta…

GL-LowPopArt: A Nearly Instance-Wise Minimax Estimator for Generalized Low-Rank Trace Regression
We present `GL-LowPopArt`, a novel Catoni-style estimator for generalized low-rank trace regression. Building on `LowPopArt` (Jang et al., 2024), it employs a two-stage approach: nuclear norm regularization followed by matrix Catoni estimation. We establish state-of-the-art estimation error bounds, surpassing existing guarantees (Fan et al., 2019; Kang et al., 2022), and reveal a novel experimental design objective, $\mathrm{GL}(π)$. The key technical challenge is controlling bias from the non…

This https://arxiv.org/abs/2202.02835 has been replaced.
link: https://scholar.google.com/scholar?q=a

Electron-phonon interaction and longitudinal-transverse phonon splitting in doped semiconductors
We study the effect of doping on the electron-phonon interaction and on the phonon frequencies in doped semiconductors, taking into account the screening in presence of free carriers at finite temperature. We study the impact of screening on the Fröhlich-like vertex and on the long-range components of the dynamical matrix, going beyond the state-of-the-art description for undoped crystals, thanks to the development of a computational method based on maximally localized Wannier functions. We ap…

Efficient Training for Optical Computing
Manon P. Bart, Nick Sparks, Ryan T. Glasser
https://arxiv.org/abs/2506.20833 https://arxiv.o…

Efficient Training for Optical Computing
Diffractive optical information processors have demonstrated significant promise in delivering high-speed, parallel, and energy efficient inference for scaling machine learning tasks. Training, however, remains a major computational bottleneck, compounded by large datasets and many simulations required for state-of-the-art classification models. The underlying linear transformations in such systems are inherently constrained to compositions of circulant and diagonal matrix factors, representing…