Retrieval Augmented Generation (RAG) uses vector databases to expand the expertise of an LLM model without having to retrain it. The idea can be applied over data lakes, leading to the notion of embedding data lakes, i.e., a pool of vector databases ready to be used by RAGs. The key component in these systems is the indexes enabling Approximated Nearest Neighbor Search (ANNS). However, in data lakes, one cannot realistically expect to build indexes for every dataset. Thus, we propose an adaptive, partition-based index, CrackIVF, that performs much better than up-front index building. CrackIVF starts answering as a small index, and only expands to improve performance as it sees enough queries. It does so by progressively adapting the index to the query workload. That way, queries can be answered right away without having to build a full index first. After seeing enough queries, CrackIVF will produce an index comparable to those built with conventional techniques. CrackIVF can often answer more than 1 million queries before other approaches have even built the index, achieving 10–1000x faster initialization times. This makes it ideal for cold or infrequently used data and as a way to bootstrap access to unseen datasets.

Methods for carefully selecting or generating a small set of training data to learn from, i.e., data pruning, coreset selection, and dataset distillation, have been shown to be effective in reducing the ever-increasing cost of training neural networks. Behind this success are rigorously designed, yet expensive, strategies for identifying the most informative training examples out of large datasets. In this work, we revisit these methods to understand if the additional computational costs associated with such strategies are justified from the perspective of time-to-accuracy, which has become a critical efficiency measure of deep neural network training over large datasets. Surprisingly, we find that many of the recently proposed methods underperform what we call Repeated Sampling of Random Subsets (RSRS or RS2), a powerful yet overlooked extension of the standard random baseline that learns from repeatedly sampled data throughout training instead of a fixed random subset. We test RS2 against thirty-two state-of-the-art data pruning and distillation methods across four datasets including ImageNet. Our results demonstrate that RS2 significantly reduces time-to-accuracy, particularly in practical regimes where accuracy, but not runtime, is similar to that of training on full dataset. For example, when training ResNet-18 on ImageNet, with 10\% of the dataset each epoch RS2 reaches an accuracy of 66\% versus 69\% when training with the full dataset. The best competing method achieves only 55\% while training 1.6$\times$ slower than RS2. Beyond the above meta-study, we discuss the theoretical properties of RS2 such as its convergence rate and generalization error. Our primary goal is to highlight that future works that aim to minimize total training cost by using subset selection, need to consider 1) the total computation cost (including preparing the subset) and 2) should aim to outperform a simple extension of random sampling (i.e., RS2).

Analytics on modern data analytic and data ware-house systems often need to run large complex queries on increasingly complex database schemas. A lot of progress has been made on executing such complex queries using techniques like scale out query processing, hardware accelerators like GPUs and code generation techniques. However, optimization of such queries remains a challenge. Existing optimal solutions either cannot be effectively parallelized, or are inefficient while doing a lot of unnecessary work. In this demonstration, we present our system, GPU-QO, which aims to demonstrate query optimization techniques for large analytical queries using GPUs. We first demonstrate Massively Parallel Dynamic Programming (MPDP) - a novel query optimization technique that can run on GPUs to generate optimal plans in a (massively) parallel and efficient manner. We then showcase IDP2-MPDP and UnionDP - two heuristic techniques, again using GPUs, that can even optimize queries containing 1000s of joins. Furthermore, we compare our techniques with current state-of-the-art solutions, and demonstrate how our techniques can reduce optimization time for optimal solutions by nearly two orders of magnitude and produce much better query plans for heuristics (up to 7x).

Modern data analytical workloads often need to run queries over a large number of tables. An optimal query plan for such queries is crucial for being able to run these queries within acceptable time bounds. However, with queries involving many tables, finding the optimal join order becomes a bottleneck in query optimization. Due to the exponential nature of join order optimization, optimizers resort to heuristic solutions after a threshold number of tables. Our objective is two fold: (a) reduce the optimization time for generating optimal plans; and (b) improve the quality of the heuristic solution.