arXiv subjects

Benito Geordie

Publications and source records attributed to Benito Geordie.

4 recordsLinked to original sources

arXiv2023

Adaptive Sampling for Deep Learning via Efficient Nonparametric Proxies

Data sampling is an effective method to improve the training speed of neural networks, with recent results demonstrating that it can even break the neural scaling laws. These results critically rely on high-quality scores to estimate the importance of an input to the network. We observe that there are two dominant strategies: static sampling, where the scores are determined before training, and dynamic sampling, where the scores can depend on the model weights. Static algorithms are computationally inexpensive but less effective than their dynamic counterparts, which can cause end-to-end slowdown due to their need to explicitly compute losses. To address this problem, we propose a novel sampling distribution based on nonparametric kernel regression that learns an effective importance score as the neural network trains. However, nonparametric regression models are too computationally expensive to accelerate end-to-end training. Therefore, we develop an efficient sketch-based approximation to the Nadaraya-Watson estimator. Using recent techniques from high-dimensional statistics and randomized algorithms, we prove that our Nadaraya-Watson sketch approximates the estimator with exponential convergence guarantees. Our sampling algorithm outperforms the baseline in terms of wall-clock time and accuracy on four datasets.

cs.LG↗

arXiv2023

Collabs: A Flexible and Performant CRDT Collaboration Framework

A collaboration framework is a distributed system that serves as the data layer for a collaborative app. Conflict-free Replicated Data Types (CRDTs) are a promising theoretical technique for implementing collaboration frameworks. However, existing frameworks are inflexible: they are often one-off implementations of research papers or only permit a restricted set of CRDT semantics, and they do not allow app-specific optimizations. Until now, there was no general framework that lets programmers mix, match, and modify CRDTs. We solve this with Collabs, a CRDT-based collaboration framework that lets programmers implement their own CRDTs, either from-scratch or by composing existing building blocks. Collabs prioritizes both semantic flexibility and performance flexibility: it allows arbitrary app-specific CRDT behaviors and optimizations, while still providing strong eventual consistency. We demonstrate Collabs's capabilities and programming model with example apps and CRDT implementations. We then show that a collaborative rich-text editor using Collabs's built-in CRDTs can scale to over 100 simultaneous users, unlike existing CRDT frameworks and Google Docs. Collabs also has lower end-to-end latency and server CPU usage than a popular Operational Transformation framework, with acceptable CRDT metadata overhead.

cs.DC↗

arXiv2023

BOLT: An Automated Deep Learning Framework for Training and Deploying Large-Scale Search and Recommendation Models on Commodity CPU Hardware

Efficient large-scale neural network training and inference on commodity CPU hardware is of immense practical significance in democratizing deep learning (DL) capabilities. Presently, the process of training massive models consisting of hundreds of millions to billions of parameters requires the extensive use of specialized hardware accelerators, such as GPUs, which are only accessible to a limited number of institutions with considerable financial resources. Moreover, there is often an alarming carbon footprint associated with training and deploying these models. In this paper, we take a step towards addressing these challenges by introducing BOLT, a sparse deep learning library for training large-scale search and recommendation models on standard CPU hardware. BOLT provides a flexible, high-level API for constructing models that will be familiar to users of existing popular DL frameworks. By automatically tuning specialized hyperparameters, BOLT also abstracts away the algorithmic details of sparse network training. We evaluate BOLT on a number of information retrieval tasks including product recommendations, text classification, graph neural networks, and personalization. We find that our proposed system achieves competitive performance with state-of-the-art techniques at a fraction of the cost and energy consumption and an order-of-magnitude faster inference time. BOLT has also been successfully deployed by multiple businesses to address critical problems, and we highlight one customer case study in the field of e-commerce.

cs.LG↗

arXiv2023

For-Each Operations in Collaborative Apps

Conflict-free Replicated Data Types (CRDTs) allow collaborative access to an app's data. We describe a novel CRDT operation, for-each on the list of CRDTs, and demonstrate its use in collaborative apps. Our for-each operation applies a given mutation to each element of a list, including elements inserted concurrently. This often preserves user intention in a way that would otherwise require custom CRDT algorithms. We give example applications of our for-each operation to collaborative rich-text, recipe, and slideshow editors.

cs.DC↗