NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval

$k$-Nearest Neighbor search on dense vector embeddings ($k$-NN retrieval) from pre-trained embedding models is the predominant retrieval method for text and images, as well as Retrieval-Augmented Generation (RAG) pipelines. In practice, application developers often fine-tune the embeddings to improve their accuracy on the dataset and query workload in hand. Existing approaches either fine-tune the pre-trained model itself or, more efficiently, but at the cost of accuracy, train adaptor models to transform the output of the pre-trained model. We present NUDGE, a family of novel non-parametric embedding fine-tuning approaches that are significantly more accurate and efficient than both sets of existing approaches. NUDGE directly modifies the embeddings of data records to maximize the accuracy of $k$-NN retrieval. We present a thorough theoretical and experimental study of NUDGE's non-parametric approach. We show that even though the underlying problem is NP-Hard, constrained variations can be solved efficiently. These constraints additionally ensure that the changes to the embeddings are modest, avoiding large distortions to the semantics learned during pre-training. In experiments across five pre-trained models and nine standard text and image retrieval datasets, NUDGE runs in minutes and often improves NDCG@10 by more than 10% over existing fine-tuning methods. On average, NUDGE provides 3.3x and 4.3x higher increase in accuracy and runs 200x and 3x faster, respectively, over fine-tuning the pre-trained model and training adaptors.

Production Machine Learning Pipelines: Empirical Analysis and Optimization Opportunities

Machine learning (ML) is now commonplace, powering data-driven applications in various organizations. Unlike the traditional perception of ML in research, ML production pipelines are complex, with many interlocking analytical components beyond training, whose sub-parts are often run multiple times on overlapping subsets of data. However, there is a lack of quantitative evidence regarding the lifespan, architecture, frequency, and complexity of these pipelines to understand how data management research can be used to make them more efficient, effective, robust, and reproducible. To that end, we analyze the provenance graphs of 3000 production ML pipelines at Google, comprising over 450,000 models trained, spanning a period of over four months, in an effort to understand the complexity and challenges underlying production ML. Our analysis reveals the characteristics, components, and topologies of typical industry-strength ML pipelines at various granularities. Along the way, we introduce a specialized data model for representing and reasoning about repeatedly run components in these ML pipelines, which we call model graphlets. We identify several rich opportunities for optimization, leveraging traditional data management ideas. We show how targeting even one of these opportunities, i.e., identifying and pruning wasted computation that does not translate to model deployment, can reduce wasted computation cost by 50% without compromising the model deployment cadence.

Whither AutoML? Understanding the Role of Automation in Machine Learning Workflows

Efforts to make machine learning more widely accessible have led to a rapid increase in Auto-ML tools that aim to automate the process of training and deploying machine learning. To understand how Auto-ML tools are used in practice today, we performed a qualitative study with participants ranging from novice hobbyists to industry researchers who use Auto-ML tools. We present insights into the benefits and deficiencies of existing tools, as well as the respective roles of the human and automation in ML workflows. Finally, we discuss design implications for the future of Auto-ML tool development. We argue that instead of full automation being the ultimate goal of Auto-ML, designers of these tools should focus on supporting a partnership between the user and the Auto-ML tool. This means that a range of Auto-ML tools will need to be developed to support varying user goals such as simplicity, reproducibility, and reliability.

Demystifying a Dark Art: Understanding Real-World Machine Learning Model Development

It is well-known that the process of developing machine learning (ML) workflows is a dark-art; even experts struggle to find an optimal workflow leading to a high accuracy model. Users currently rely on empirical trial-and-error to obtain their own set of battle-tested guidelines to inform their modeling decisions. In this study, we aim to demystify this dark art by understanding how people iterate on ML workflows in practice. We analyze over 475k user-generated workflows on OpenML, an open-source platform for tracking and sharing ML workflows. We find that users often adopt a manual, automated, or mixed approach when iterating on their workflows. We observe that manual approaches result in fewer wasted iterations compared to automated approaches. Yet, automated approaches often involve more preprocessing and hyperparameter options explored, resulting in higher performance overall--suggesting potential benefits for a human-in-the-loop ML system that appropriately recommends a clever combination of the two strategies.

Helix: Holistic Optimization for Accelerating Iterative Machine Learning

Machine learning workflow development is a process of trial-and-error: developers iterate on workflows by testing out small modifications until the desired accuracy is achieved. Unfortunately, existing machine learning systems focus narrowly on model training---a small fraction of the overall development time---and neglect to address iterative development. We propose Helix, a machine learning system that optimizes the execution across iterations---intelligently caching and reusing, or recomputing intermediates as appropriate. Helix captures a wide variety of application needs within its Scala DSL, with succinct syntax defining unified processes for data preprocessing, model specification, and learning. We demonstrate that the reuse problem can be cast as a Max-Flow problem, while the caching problem is NP-Hard. We develop effective lightweight heuristics for the latter. Empirical evaluation shows that Helix is not only able to handle a wide variety of use cases in one unified workflow but also much faster, providing run time reductions of up to 19x over state-of-the-art systems, such as DeepDive or KeystoneML, on four real-world applications in natural language processing, computer vision, social and natural sciences.

The Importance of Norm Regularization in Linear Graph Embedding: Theoretical Analysis and Empirical Demonstration

Learning distributed representations for nodes in graphs is a crucial primitive in network analysis with a wide spectrum of applications. Linear graph embedding methods learn such representations by optimizing the likelihood of both positive and negative edges while constraining the dimension of the embedding vectors. We argue that the generalization performance of these methods is not due to the dimensionality constraint as commonly believed, but rather the small norm of embedding vectors. Both theoretical and empirical evidence are provided to support this argument: (a) we prove that the generalization error of these methods can be bounded by limiting the norm of vectors, regardless of the embedding dimension; (b) we show that the generalization performance of linear graph embedding methods is correlated with the norm of embedding vectors, which is small due to the early stopping of SGD and the vanishing gradients. We performed extensive experiments to validate our analysis and showcased the importance of proper norm regularization in practice.

Helix: Accelerating Human-in-the-loop Machine Learning

Data application developers and data scientists spend an inordinate amount of time iterating on machine learning (ML) workflows -- by modifying the data pre-processing, model training, and post-processing steps -- via trial-and-error to achieve the desired model performance. Existing work on accelerating machine learning focuses on speeding up one-shot execution of workflows, failing to address the incremental and dynamic nature of typical ML development. We propose Helix, a declarative machine learning system that accelerates iterative development by optimizing workflow execution end-to-end and across iterations. Helix minimizes the runtime per iteration via program analysis and intelligent reuse of previous results, which are selectively materialized -- trading off the cost of materialization for potential future benefits -- to speed up future iterations. Additionally, Helix offers a graphical interface to visualize workflow DAGs and compare versions to facilitate iterative development. Through two ML applications, in classification and in structured prediction, attendees will experience the succinctness of Helix programming interface and the speed and ease of iterative development using Helix. In our evaluations, Helix achieved up to an order of magnitude reduction in cumulative run time compared to state-of-the-art machine learning tools.

How Developers Iterate on Machine Learning Workflows -- A Survey of the Applied Machine Learning Literature

Machine learning workflow development is anecdotally regarded to be an iterative process of trial-and-error with humans-in-the-loop. However, we are not aware of quantitative evidence corroborating this popular belief. A quantitative characterization of iteration can serve as a benchmark for machine learning workflow development in practice, and can aid the development of human-in-the-loop machine learning systems. To this end, we conduct a small-scale survey of the applied machine learning literature from five distinct application domains. We collect and distill statistics on the role of iteration within machine learning workflow development, and report preliminary trends and insights from our investigation, as a starting point towards this benchmark. Based on our findings, we finally describe desiderata for effective and versatile human-in-the-loop machine learning systems that can cater to users in diverse domains.

On the Interpretability of Conditional Probability Estimates in the Agnostic Setting

We study the interpretability of conditional probability estimates for binary classification under the agnostic setting or scenario. Under the agnostic setting, conditional probability estimates do not necessarily reflect the true conditional probabilities. Instead, they have a certain calibration property: among all data points that the classifier has predicted P(Y = 1|X) = p, p portion of them actually have label Y = 1. For cost-sensitive decision problems, this calibration property provides adequate support for us to use Bayes Decision Theory. In this paper, we define a novel measure for the calibration property together with its empirical counterpart, and prove an uniform convergence result between them. This new measure enables us to formally justify the calibration property of conditional probability estimations, and provides new insights on the problem of estimating and calibrating conditional probabilities.

Indexing Cost Sensitive Prediction

Predictive models are often used for real-time decision making. However, typical machine learning techniques ignore feature evaluation cost, and focus solely on the accuracy of the machine learning models obtained utilizing all the features available. We develop algorithms and indexes to support cost-sensitive prediction, i.e., making decisions using machine learning models taking feature evaluation cost into account. Given an item and a online computation cost (i.e., time) budget, we present two approaches to return an appropriately chosen machine learning model that will run within the specified time on the given item. The first approach returns the optimal machine learning model, i.e., one with the highest accuracy, that runs within the specified time, but requires significant up-front precomputation time. The second approach returns a possibly sub- optimal machine learning model, but requires little up-front precomputation time. We study these two algorithms in detail and characterize the scenarios (using real and synthetic data) in which each performs well. Unlike prior work that focuses on a narrow domain or a specific algorithm, our techniques are very general: they apply to any cost-sensitive prediction scenario on any machine learning algorithm.