the archive

#generalization

8 results

cs.LG2019

Adaptive Weight Decay for Deep Neural Networks

Kensuke Nakamura, Byung-Woo Hong

The paper proposes AdaDecay, an adaptive weight-decay technique that adjusts regularization per parameter based on gradient magnitude, and demonstrates improved generalization on M…

#regularization#weight decay#adaptive optimization#deep neural networks

stat.ML2019

Robust Learning with Jacobian Regularization

Judy Hoffman, Daniel A. Roberts, Sho Yaida

The paper proposes an efficient Jacobian regularization method for neural networks that enlarges classification margins and improves robustness to both random and adversarial input…

#robustness#adversarial training#jacobian regularization#neural networks

cs.LG2019

Learning to Generalize to Unseen Tasks with Bilevel Optimization

Hayeon Lee, Donghyun Na, Hae Beom Lee +1

The paper introduces L2G, a bilevel optimization framework that explicitly encourages metric‑based meta‑learning models to generalize to unseen classification tasks, leading to imp…

#meta-learning#few-shot classification#metric learning#bilevel optimization

math.GM2019

Further Generalization of Golden Mean in Relation to Euler Divine Equation

Miloje M. Rakocevic

The paper proposes a new generalization of the golden mean and observes that Euler's divine equation can serve as another possible generalization of the golden ratio.

#golden ratio#generalization#euler equation#number theory

cs.LG2019

Learning Overparameterized Neural Networks via Stochastic Gradient Descent on Structured Data

Yuanzhi Li, Yingyu Liang

The paper analyzes how stochastic gradient descent can successfully train an overparameterized two‑layer ReLU network for multi‑class classification when the data come from well‑se…

#overparameterization#stochastic gradient descent#two-layer neural networks#ReLU activation

cs.CV2019

Safe Augmentation: Learning Task-Specific Transformations from Data

Irynei Baran, Orest Kupyn, Arseny Kravchenko

The paper proposes Safe Augmentation, a simple, model‑agnostic approach that learns task‑specific image transformations which preserve the data distribution and boost generalizatio…

#data augmentation#task-specific transformations#model fine-tuning#generalization

cs.CV2019

Meta Learning for Task-Driven Video Summarization

Xuelong Li, Hongli Li, Yongsheng Dong

The paper introduces MetaL‑TDVS, a meta‑learning framework that treats the summarization of each video as a separate task, enabling the model to leverage experience from other vide…

#meta learning#video summarization#task-driven learning#generalization

stat.ML2019

Generalisation in fully-connected neural networks for time series forecasting

Anastasia Borovykh, Cornelis W. Oosterlee, Sander M. Bohte

The paper examines how fully‑connected neural networks generalize in time‑series forecasting by measuring input and weight Hessians, and shows that training hyperparameters like le…

#time series forecasting#generalization#fully-connected networks#hessian analysis