22k citations · 26.7k across the 60 of their papers we have counts for
19 papers · 1 filter
Compression Scaling Laws:Unifying Sparsity and Quantization
Elias Frantar, Utku Evci, Wonpyo Park +2
We investigate how different compression techniques -- such as weight and activation quantization, and weight sparsity -- affect the scaling behavior of large language models (LLMs…
Scaling Laws for Sparsely-Connected Foundation Models
Elias Frantar, Carlos Riquelme, Neil Houlsby +2
We explore the impact of parameter sparsity on the scaling behavior of Transformers trained on massive datasets (i.e., "foundation models"), in both vision and language domains. In…
From Sparse to Soft Mixtures of Experts
Joan Puigcerver, Carlos Riquelme, Basil Mustafa +1
Sparse mixture of expert architectures (MoEs) scale model capacity without significant increases in training or inference costs. Despite their success, MoEs suffer from a number of…
Massively Scaling Heteroscedastic Classifiers
Mark Collier, Rodolphe Jenatton, Basil Mustafa +3
Heteroscedastic classifiers, which learn a multivariate Gaussian distribution over prediction logits, have been shown to perform well on image classification problems with hundreds…
Adaptive Computation with Elastic Input Sequence
Fuzhao Xue, Valerii Likhosherstov, Anurag Arnab +3
Humans have the ability to adapt the type of information they use, the procedure they employ, and the amount of time they spend when solving problems. However, most standard neural…
Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints
Aran Komatsuzaki, Joan Puigcerver, James Lee-Thorp +6
Training large, deep neural networks to convergence can be prohibitively expensive. As a result, often only a small selection of popular, dense models are reused across different c…
Sparse MoEs meet Efficient Ensembles
James Urquhart Allingham, Florian Wenzel, Zelda E Mariet +10
Machine learning models based on the aggregated outputs of submodels, either at the activation or prediction levels, often exhibit strong performance compared to individual models.…
The Benchmark Lottery
Mostafa Dehghani, Yi Tay, Alexey A. Gritsenko +5
The world of empirical machine learning (ML) strongly relies on benchmarks in order to determine the relative effectiveness of different algorithms and methods. This paper proposes…
Revisiting the Calibration of Modern Neural Networks
Matthias Minderer, Josip Djolonga, Rob Romijnders +5
Accurate estimation of predictive uncertainty (model calibration) is essential for the safe application of neural networks. Many instances of miscalibration in modern neural networ…
Comparing Transfer and Meta Learning Approaches on a Unified Few-Shot Classification Benchmark
Vincent Dumoulin, Neil Houlsby, Utku Evci +4
Meta and transfer learning are two successful families of approaches to few-shot learning. Despite highly related goals, state-of-the-art advances in each family are measured large…
Underspecification Presents Challenges for Credibility in Modern Machine Learning
Alexander D'Amour, Katherine Heller, Dan Moldovan +37
ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline i…
Deep Ensembles for Low-Data Transfer Learning
Basil Mustafa, Carlos Riquelme, Joan Puigcerver +3
In the low-data regime, it is difficult to train good supervised models from scratch. Instead practitioners turn to pre-trained models, leveraging transfer learning. Ensembling is…
Scalable Transfer Learning with Expert Models
Joan Puigcerver, Carlos Riquelme, Basil Mustafa +5
Transfer of pre-trained representations can improve sample efficiency and reduce computational requirements for new tasks. However, representations used for transfer are usually ge…
Parameter-Efficient Transfer Learning for NLP
Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski +5
Fine-tuning large pre-trained models is an effective transfer mechanism in NLP. However, in the presence of many downstream tasks, fine-tuning is parameter inefficient: an entire n…
Neural Architecture Search Over a Graph Search Space
Stanisław Jastrzębski, Quentin de Laroussilhe, Mingxing Tan +3
Neural Architecture Search (NAS) enabled the discovery of state-of-the-art architectures in many domains. However, the success of NAS depends on the definition of the search space.…
Self-Supervised GAN to Counter Forgetting
Ting Chen, Xiaohua Zhai, Neil Houlsby
GANs involve training two networks in an adversarial game, where each network's task depends on its adversary. Recently, several works have framed GAN training as an online or cont…
Self-Supervised GANs via Auxiliary Rotation Loss
Ting Chen, Xiaohua Zhai, Marvin Ritter +2
Conditional GANs are at the forefront of natural image synthesis. The main drawback of such models is the necessity for labeled data. In this work we exploit two popular unsupervis…
On Self Modulation for Generative Adversarial Networks
Ting Chen, Mario Lucic, Neil Houlsby +1
Training Generative Adversarial Networks (GANs) is notoriously challenging. We propose and study an architectural modification, self-modulation, which improves GAN performance acro…
Transfer Learning with Neural AutoML
Catherine Wong, Neil Houlsby, Yifeng Lu +1
We reduce the computational cost of Neural AutoML with transfer learning. AutoML relieves human effort by automating the design of ML algorithms. Neural AutoML has become popular f…