Zobrazeno 1 - 10
of 29 918
pro vyhledávání: '"Nikhil, P. P."'
While scaling laws provide a reliable methodology for predicting train loss across compute scales for a single data distribution, less is known about how these predictions should change as we change the distribution. In this paper, we derive a strate
Externí odkaz:
http://arxiv.org/abs/2411.12925
Autor:
Jelassi, Samy, Mohri, Clara, Brandfonbrener, David, Gu, Alex, Vyas, Nikhil, Anand, Nikhil, Alvarez-Melis, David, Li, Yuanzhi, Kakade, Sham M., Malach, Eran
The Mixture-of-Experts (MoE) architecture enables a significant increase in the total number of model parameters with minimal computational overhead. However, it is not clear what performance tradeoffs, if any, exist between MoEs and standard dense t
Externí odkaz:
http://arxiv.org/abs/2410.19034
The eigenvalues and eigenvectors of nonnormal matrices can be unstable under perturbations of their entries. This renders an obstacle to the analysis of numerical algorithms for non-Hermitian eigenvalue problems. A recent technique to handle this iss
Externí odkaz:
http://arxiv.org/abs/2411.19926
Autor:
Gabellini, Cristian, Shenoy, Nikhil, Thaler, Stephan, Canturk, Semih, McNeela, Daniel, Beaini, Dominique, Bronstein, Michael, Tossou, Prudencio
Machine Learning Interatomic Potentials (MLIPs) are a highly promising alternative to force-fields for molecular dynamics (MD) simulations, offering precise and rapid energy and force calculations. However, Quantum-Mechanical (QM) datasets, crucial f
Externí odkaz:
http://arxiv.org/abs/2411.19629
Autor:
Behari, Nikhil, Young, Aaron, Somasundaram, Siddharth, Klinghoffer, Tzofi, Dave, Akshat, Raskar, Ramesh
3D surface reconstruction is essential across applications of virtual reality, robotics, and mobile scanning. However, RGB-based reconstruction often fails in low-texture, low-light, and low-albedo scenes. Handheld LiDARs, now common on mobile device
Externí odkaz:
http://arxiv.org/abs/2411.19474
Autor:
Udandarao, Vishaal, Parthasarathy, Nikhil, Naeem, Muhammad Ferjad, Evans, Talfan, Albanie, Samuel, Tombari, Federico, Xian, Yongqin, Tonioni, Alessio, Hénaff, Olivier J.
Knowledge distillation (KD) is the de facto standard for compressing large-scale models into smaller ones. Prior works have explored ever more complex KD strategies involving different objective functions, teacher-ensembles, and weight inheritance. I
Externí odkaz:
http://arxiv.org/abs/2411.18674
The influence of contextual input on the behavior of large language models (LLMs) has prompted the development of context attribution methods that aim to quantify each context span's effect on an LLM's generations. The leave-one-out (LOO) error, whic
Externí odkaz:
http://arxiv.org/abs/2411.15102
The gold standard in human-AI collaboration is complementarity -- when combined performance exceeds both the human and algorithm alone. We investigate this challenge in binary classification settings where the goal is to maximize 0-1 accuracy. Given
Externí odkaz:
http://arxiv.org/abs/2411.15230
Autor:
Mishra, Nikhil
This work addresses a trust-based enhancement to the Multipath Ad hoc On-Demand Distance Vector (AOMDV) routing protocol. While AODV and its multipath variant AOMDV have been fundamental in mobile ad hoc networks, they lack mechanisms to account for
Externí odkaz:
http://arxiv.org/abs/2411.13227
Galaxies grow and evolve in dark matter halos. Because dark matter is not visible, galaxies' halo masses ($\rm{M}_{\rm{halo}}$) must be inferred indirectly. We present a graph neural network (GNN) model for predicting $\rm{M}_{\rm{halo}}$ from stella
Externí odkaz:
http://arxiv.org/abs/2411.12629