Reconciling Differences in Scaling Laws Between Kaplan and Chinchilla Studies
The core message of this paper is that much of the discrepancy between the scaling coefficient estimates reported in the Kaplan and Chinchilla studies can be attributed to Kaplan counting non-embedding rather than total parameters, combined with their analysis being performed at small scale.
삼각형 및 테트라헤드론에 대한 심플리셜 텐서 유한요소의 공식 및 변환
폴리토프 템플릿 방법을 사용하여 다양한 부분적으로 연속적인 텐서 값 유한요소 공간을 통일된 방식으로 구축할 수 있다.
単体テンソル有限要素の多角形テンプレートを用いた公式と変換
本論文では、多角形テンプレートを用いて、様々な部分的に連続なテンソル値有限要素空間を単体上に構築する統一的な手法を提案する。
Unified Construction of Tensor-Valued Finite Elements on Simplices Using Polytopal Templates
This work introduces a unified method for constructing the basis functions of a wide variety of partially continuous tensor-valued finite elements on simplices using polytopal templates. The proposed approach allows for the construction of well-known elements such as Regge, Hellan-Herrmann-Johnson, Pechstein-Schöberl, Hu-Zhang, Hu-Ma-Sun, and Gopalakrishnan-Lederer-Schöberl elements.
분자 표현의 계층적 구조가 선형 분자 표현의 일반화 능력을 향상시킨다: t-SMILES 2
분자 표현의 계층적 구조가 선형 분자 표현의 일반화 능력을 향상시킨다.
High-Resolution 3D Cine MRI Analysis of Left Atrial Displacements and Strains using Online Learning Neural Networks
Aladdin, a novel end-to-end workflow, can reliably provide high-resolution 3D maps of left atrial displacements and strains from cardiac cine MRI, enabling comprehensive characterization of left atrial function.
반정밀도 파동 시뮬레이션의 정확성 향상
반정밀도 파동 시뮬레이션에서 발생하는 정확도 저하 문제를 보상 합계 기법을 통해 해결할 수 있다.
Mitigating Roundoff Errors in Half Precision Wave Simulations through Compensated Summation
Applying compensated summation can greatly restore the solution quality for half precision wave simulations by addressing the accumulation of roundoff errors in the solution updates.
Optimal Density Estimation under Central Privacy Constraints
The core message of this article is that the cost of central privacy in density estimation depends on the smoothness of the underlying density and the privacy budget. For Lipschitz densities, the minimax rate of estimation is degraded when the privacy budget is small, but for smoother Sobolev densities, the minimax rate can be preserved in certain privacy regimes.
Architectural Biases in Untrained Neural Networks: How Model Design Choices Can Skew Initial Predictions
Architectural choices in neural network design, such as activation functions, pooling layers, and data preprocessing, can introduce an inherent bias in the initial predictions of untrained models, even in the absence of explicit biases in the data or training process.
© 2024 by Linnk AI