ALF: An Active Learning Framework for Scientific Discovery

Shikha Surana; Alex Hawkins-Hooker; Olivia Gallup; Christoph Brunken; Jules Tilly; Paul Duckworth

cs.LGSep 25, 2026

ALF: An Active Learning Framework for Scientific Discovery

Authors: Shikha Surana, Alex Hawkins-Hooker, Olivia Gallup, Christoph Brunken, Jules Tilly, Paul Duckworth

Organizations: InstaDeep · University of Oxford

Abstract

Machine learning for scientific discovery is almost systematically data bound. Producing relevant high quality data, under budget constraints, is amongst the most promising ways to advance the field. Active learning (AL) offers promise wherever labelling requires expensive experiment, measurement, or simulation. Most existing tools cover only part of the data acquisition loop, and typically focus on either offline benchmarking or online deployment, but not both. We present ALF, a modular AL Framework that runs the full data acquisition loop via five modular components. One clear API for both settings: offline, against an existing dataset for controlled and reproducible experimentation; and online, against an oracle for acquiring new candidates in real-world deployments. ALF is open-source and available at https://github.com/instadeepai/alf.

Figures & tables

Figure 1: ALF: Data acquisition loop. At each round, a surrogate model is first fit then queried (ask) to iteratively approximate an oracle of interest.

Figure 3: Small molecule design . Offline benchmarking of Active learning objective (left) and Bayesian optimisation (right). Results are the mean over

5

random seeds; error bars are

\pm 1

std.

Figure 4: Small molecule design . Online data acquisition for Active Learning objective (left) and Bayesian optimisation (right), against an RDKit oracle, with single-mutation search. Results are the mean over

5

random seeds; error bars are

\pm 1

std.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Component	You implement	Provided by the base
Dataset	load_dataset	splitting, query , split updates, metrics
Model	featurise , train ,	setup , embedding hook, metric
	predict , sample	collection; reuse across roles
Acquisition fn.	call(candidates, state)	top- $k$ batching, optimizer integration
Search fn.	call(state)	dataset / generator / protocol subclasses
Oracle	— (wrap a dataset or model)	offline/online branching

Appendix

Table 1: What each component requires from a subclass, and what the base class provides. Oracle and Surrogate are configured rather than subclassed.

Figure 5: Acquisition strategies across sequence, molecule, and materials under two objectives, from single ALF campaigns. Top, active learning: test-set RMSE against acquisition round (lower is better) for the information-based strategies. Bottom, Bayesian optimisation: best score acquired so far (higher is better) for the utility-based strategies. Random is shown in both rows as the reference. Mean of five seeds,

\pm 1

s.d. bands.

Figure 6: Acquisition strategies across sequence, molecule, and materials under two objectives, from single online ALF campaigns. Top, active learning: to-be acquired batch RMSE against acquisition round (lower is better) for the information-based strategies. Bottom, Bayesian optimisation: best score acquired so far (higher is better) for the utility-based strategies. Random is shown in both rows as the reference. Mean of five seeds,

\pm 1

s.d. bands.

Panel of Figure 7 — Figure 7: ALF in published research. (a) Active learning for interatomic potentials Varga-Umbrich et al. [2026a] , Varga-Umbrich et al. [2026b] : force error per acquisition round, with a new neural-tangent-kernel signal among the strategies. (b, c) Bayesian optimisation for protein design [ Hawkins-Hooker et al., 2026 ] : a ranking-based fine-tuning loss recovers more top candidates per round than MSE (b), and ProteinNPT [ Notin et al., 2023 ] and other surrogates are compared in the same multi-round design setting (c).

Active Learning

Active learning selects the examples, labels, features, or experiments most useful for improving a model under limited budgets.

See all 280 papers

LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs
May 21, 2026Sanchit Kabra, Nikhil Abhyankar, Saaketh Desai +2Scientific Discovery Scientific Hypothesis Generation
1Virginia Tech · 2Sandia National Laboratories
Soft Strategy Selection for Batch-Mode Active Learning
Oct 4, 2026Rushil Gupta, Romain LopezActive Learning
Department of Computer Science, Courant Institute School of Mathematics, Computing, and Data Science, New York University · Department of Biology, New York University
Foundation-Model Surrogates Enable Data-Efficient Active Learning for Materials Discovery
Mar 13, 2026Jeffrey Hu, Rongzhi Dong, Ying Feng +2Active Learning Data-Efficient Learning
Department of Materials Science and Engineering, University of Illinois Urbana Champaign, Champaign,IL, USA · Department of Computer Science & Engineering, University of South Carolina, Columbia, SC, USA · Zuoyue Honors College, Hangzhou Dianzhi University, Hangzhou, China +1

May 21, 2026J/K move · Enter open · S save

LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs

Sanchit Kabra, Nikhil Abhyankar, Saaketh Desai, Prasad Iyer+1

1Virginia Tech · 2Sandia National Laboratories

Scientific Discovery Scientific Hypothesis Generation Large Language Model-Guided Scientific Discovery

Open paper AskPDF arXiv

Scientific discovery is a closed-loop process in which hypotheses guide data acquisition and observations refine the hypothesis space. Yet most approaches reduce discovery to supervised learning over fixed datasets, where limited observations can support multiple plausible mechanisms that fit locally but fail to generalize. Thus, the key challenge is selecting informative observations to resolve uncertainty, shifting the focus from static inference to adaptive data acquisition. To address this, we propose LLM-AutoSciLab, a closed-loop framework that couples hypothesis generation with hypothesis-conditioned experiment selection and mechanism refinement. Rather than fitting models to passively collected data, LLM-AutoSciLab iteratively proposes plausible hypotheses, selects informative experiments to distinguish or refine them, and updates its state using the resulting evidence. To evaluate dynamic, closed-loop scientific discovery with active data acquisition, we introduce ActiveSciBench, comprising two datasets: ActiveSciBench-Chem with 57 enzyme-kinetics tasks and ActiveSciBench-GRN with 45 gene-regulatory-network tasks. These datasets model discovery as a budget-constrained process requiring adaptive experiment design, variable selection, and recovery of true mechanisms. Across NewtonBench, ActiveSciBench-Chem, and ActiveSciBench-GRN, LLM-AutoSciLab outperforms prior methods, achieving 67.6% and 35.1% symbolic accuracy on NewtonBench and ActiveSciBench-Chem, respectively, and 31.1% exact graph recovery on ActiveSciBench-GRN. Moreover, hypothesis-guided experimentation is 2-5x more sample-efficient than the strongest competing baselines. Code and data are available at: https://github.com/scientific-discovery/LLM-AutoSciLab

Abstract

Active Learning

Explore similar work

LLM-AutoSciLab: Closed-Loop Scientific Discovery via Active Experimentation with LLMs

Soft Strategy Selection for Batch-Mode Active Learning

Foundation-Model Surrogates Enable Data-Efficient Active Learning for Materials Discovery