Pith. sign in

REVIEW 3 major objections 4 minor 27 references

Curiosity Driven Exploration to Optimize Structure-Property Learning in Microscopy

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper shows that a curiosity-driven active-learning algorithm, which samples where a surrogate model predicts the strongest imaging-to-spectroscopy errors, lowers prediction error compared with random sampling and runs in real time…

desk verdict Useful integration of error-prediction active learning with Im2spec/Spec2im, but the headline 'beats random' claim may conflate the acquisition rule with k-means initialization; fix the baseline and statistics before trusting it. read the letter →

arxiv 2504.20011 v2 pith:5DNANGDF submitted 2025-04-28 cond-mat.mtrl-sci cs.LG

Aditya Vatsavai , Ganesh Narasimha , Yongtao Liu , Jawad Chowdhury , Jan-Chi Yang , Hiroshi Funakubo , Maxim Ziatdinov , Rama Vasudevan This is my paper · ORCID
classification cond-mat.mtrl-scics.LG
keywords activelearningcuriosity-drivenexplorationstructure-propertyrelationshipspiezoresponseforcemicroscopyIm2specSpec2imdeepautonomous
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that a lightweight 'curiosity' algorithm can decide where a microscope should take its next spectroscopic measurement, and that this rule beats random sampling for learning structure-property mappings. The algorithm trains an auxiliary deep network to predict where the main image-to-spectrum (Im2spec) or spectrum-to-image (Spec2im) model is currently wrong, then measures at the spot with the highest predicted error. On pre-acquired ferroelectric data, this lowers prediction loss for a fixed measurement budget; the authors also run the loop in real time on an atomic force microscope. If true, the method gives materials scientists a cheap, deployable way to find structure-property relationships without the computational burden of Bayesian active learning.

What carries the argument

The central object is an auxiliary error-prediction network: a surrogate that learns to forecast where the main structure-property model is failing. In the first workflow, the frozen encoder of the best Im2spec model is joined to a new decoder trained on the L1 spectral-mismatch error, and the acquisition function $A_j = 1 - e^{-\lambda |L_j - (1-\beta)|}$ tunes between exploitation ($\beta=0$) and curiosity-driven exploration ($\beta=1$). In the second workflow, an autoencoder maps every image patch to a latent embedding, an error predictor is trained on normalized mean-squared-error targets, and the acquisition function combines predicted error $C_j$ with an exploration reward $E_j$—the harmonic mean of distances to already measured points in latent space—under an epsilon-decreasing schedule. The error model converts 'where is the model ignorant?' into a measurable score that chooses the next spectrum.

What would settle it

Run the 200-iteration autoencoder workflow many times with identical initialization but with the error predictor's training labels shuffled relative to the image patches, so predicted error carries no information about true Im2spec error. If the shuffled-curiosity acquisition still matches or beats random sampling, the reported gain is not caused by error prediction; conversely, a dataset in which the highest predicted-error regions are pure noise would exhibit the plateau failure mode the authors acknowledge, showing the claim depends on informative regions coinciding with high error.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a neural-network surrogate trained to predict where the structure-property model currently fails provides a usable acquisition signal for active learning in microscopy. Against random sampling on pre-acquired piezoresponse force microscopy data, the curiosity algorithm reaches a lower minimum loss with the same number of measurements, and the difference is statistically significant. The authors further find that adding Monte Carlo dropout uncertainty to the exploration reward does not lower average Im2spec loss but does reduce error on the ten hardest points, and that the loop can be deployed on a live microscope in both forward and inverse directions. They draw the corollary that a single curiosity policy cannot optimize both directions at once, because several structures can produce the same hysteresis loop.

Load-bearing premise

The load-bearing assumption is that the spot where the model currently expects to be wrong is the most valuable place to measure next, and that the auxiliary error predictor can keep learning a target that changes as the main model improves.

Editorial extensions

If this is right

  • With the same budget of spectroscopic measurements, the curiosity algorithm reaches a lower test-set prediction loss than random sampling; in the autoencoder benchmark over 100 trials the difference is statistically significant.
  • Acquisitions concentrate on domain walls and defective regions where structure-to-spectrum correlation is hardest, so the error signal identifies physically informative sites instead of filling a regular grid.
  • The loop runs in real time on a microscope, so both Im2spec and Spec2im versions are deployable during an experiment.
  • Because the structure–spectrum correspondence is not one-to-one, a single curiosity policy cannot simultaneously optimize the forward and inverse problems; the appropriate direction must be chosen for the application.
  • The method can stall on regions where high predicted error reflects noise rather than learnable structure, so practical use needs a plateau-escape mechanism such as human intervention or injected stochasticity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the authors leave implicit is that persistent high predicted error at the same latent locations across iterations could serve as a stopping rule or a flag for regions where no learned mapping will suffice, directing the user to a different probe, a theoretical model, or a human decision.
  • A clean way to separate the contribution of the error signal from the exploration reward would be to benchmark curiosity against pure dropout-uncertainty sampling and pure latent-distance sampling; the paper's combined acquisition function makes the individual contributions hard to isolate.
  • Because the authors show the forward and inverse acquisition paths differ on the same sample, an autonomous experiment that needs both models could alternate between the two acquisition functions or weight them by the user's downstream task rather than committing to one policy.
  • The same machinery should transfer to other microscopy modalities in which an image patch predicts a spectrum, such as electron-energy-loss or cathodoluminescence mapping, provided the patch size is chosen on physical grounds as done here for ferroelectric domains.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript presents two "curiosity-driven" active-learning workflows for learning structure-property relationships in scanning probe microscopy. In the first workflow (Sec. III.A), an ensemble of Im2spec models predicts spectra from image patches, and a frozen-encoder auxiliary network is trained to predict the L1 spectral mismatch; acquisitions are selected by an empirical function with an exploration/exploitation parameter beta. In the second workflow (Sec. III.B), an autoencoder latent representation of image patches feeds an error predictor of Im2spec MSE, with k-means initialization and a harmonic-mean exploration reward combined with predicted error. On a pre-acquired PFM dataset the authors report that the curiosity algorithm achieves lower Im2spec loss than random sampling (Fig. 6), and they deploy a related sampler on a real AFM for both Im2spec and Spec2im tasks (Sec. III.C). The paper argues that the approach is a lightweight alternative to deep kernel active learning for multidimensional, correlated spectral outputs, and it provides public code and data.

Significance. If the central claim is established, the contribution is useful: a lightweight, curiosity-style acquisition rule that can reduce the number of spectroscopic measurements needed to learn structure-property maps, with a real-time microscope demonstration and open source code and data. The paper honestly discusses known failure modes (initialization sensitivity, non-stationary error targets, plateau trapping), and the second pre-acquired experiment uses repeated trials, both of which are strengths. The main weakness is that the quantitative head-to-head evidence in Fig. 6 is under-specified: the baseline protocol is not stated to be matched, the trial count is inconsistent between text and caption, and no test statistic is reported despite the claim of statistical significance. These issues are fixable and do not invalidate the approach, but they must be addressed before the outperformance claim can be verified.

major comments (3)
  1. [§III.B, Fig. 6] The benchmark against random sampling does not yet establish that the gain comes from curiosity-driven acquisition. The text states that the algorithm is sensitive to initialization and therefore initializes via k-means in the autoencoder latent space, but it does not state whether the random baseline also received the same k-means initialization, the same Im2spec retraining schedule, and the same error-predictor training budget. If the random baseline uses uniform random starts, the lower loss in Fig. 6 could be caused by better initial coverage rather than by the acquisition rule. Please run a matched-protocol baseline (same initialization, same retraining schedule, same per-iteration compute) and report loss at a fixed measurement budget, not only the 'minimum loss achieved.'
  2. [§III.B, Fig. 6 caption] The statistical evidence for the headline claim is internally inconsistent and incomplete. The main text says the overall loss metrics were obtained 'after running 100 trials,' whereas the Fig. 6 caption says '30 trials,' and the claim that the difference between the means is statistically significant is made without reporting the test, the effect size, or a p-value. Please reconcile the trial count, report the statistical procedure and its result, and show per-iteration mean and confidence intervals so the reader can see the loss trajectories rather than only the minimum.
  3. [§III.B] The central mechanism assumes that the error predictor supplies a reliable acquisition signal, yet the paper acknowledges that the error targets are non-stationary and that training the predictor is 'very challenging,' requiring a large learning rate and multiple epochs per measurement. Because the acquisition function also contains an explicit exploration reward, the reported improvement could in principle come primarily from that reward rather than from predicted-error-based curiosity. Please add a diagnostic of error-predictor quality over iterations (e.g., correlation between predicted and actual MSE on a held-out set) or an ablation that removes the error term from the acquisition function, so the contribution of the error model to the Fig. 6 result is separable.
minor comments (4)
  1. [§V] The data availability statement misspells 'Zenodo' as 'Zenedo'; please correct it.
  2. [§III.C] The real-time deployment section describes the results qualitatively ('decent predictions'); adding a quantitative metric, such as final test loss or a comparison with a random run on the same sample, would make the demonstration easier to evaluate.
  3. [§III.A, Eq. (1)] The acquisition function in Eq. (1) would be clearer if the symbols L_j, lambda, and beta were typeset with subscripts and their normalization explained immediately after the equation; the current text leaves the reader to infer that L_j is the normalized L1 error.
  4. [§III.B, Fig. 5] Panels (d)-(f) of Fig. 5 would benefit from color bars and a sentence quantifying the spatial correlation between the predicted error map and the sampled exploration path, since the text relies on qualitative visual inspection.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical benchmark against random sampling on held-out data, not a derived quantity that reduces to its inputs.

full rationale

The paper's derivation chain is operational rather than self-referential. The workflow trains an Im2spec model, trains an auxiliary error predictor on the resulting spectral mismatch, and then uses an acquisition function to sample points with high predicted error; the claimed benefit is measured by comparing the final Im2spec loss against a random-sampling baseline under the same measurement budget. The acquisition function A_j = 1 - exp(-lambda |L_j - (1 - beta)|) is indeed designed to track predicted error, but that is an algorithm design choice, not a circular prediction. The headline result, that curiosity sampling achieves lower loss than random sampling, is an empirical statement about test-set or overall loss after active learning, and therefore could have come out differently; it is not guaranteed by the definition of the acquisition function. Self-citations in the paper are used for the im2spec codebase, the AEcroscopy platform, and prior datasets, but none of these citations supplies the load-bearing claim that curiosity sampling outperforms random sampling. That claim is benchmarked in the paper against random acquisition and reported with trial statistics. The acknowledged weaknesses, such as sensitivity to initialization, non-stationary error-predictor targets, and plateau trapping, are experimental limitations rather than signs that the result is a tautology. The k-means initialization concern raised by a skeptical reader is a matched-baseline experimental-design question, not a circularity: even if the initialization advantage partly explains the result, the comparison is still an external benchmark. No uniqueness theorem, fitted parameter renamed as a prediction, or self-citation chain is used to force the central conclusion. The paper is therefore not circular.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a set of hand-set hyperparameters (beta, lambda, patch size, latent dimension, initialization size) and on the domain assumption that image patches determine spectra. The error-prediction acquisition is an empirical construction, not derived from a formal model. No external physical constants are fitted and no entities are invented.

free parameters (6)
  • Acquisition beta (exploration/exploitation) = 0, 0.5, 1 (extreme values studied)
    Controls the trade-off between sampling high-error (beta=1) and low-error (beta=0) points; no principled selection criterion is given.
  • Acquisition lambda (smoothness prefactor) = 0.1 (first workflow); unspecified in second
    Chosen empirically; the acquisition function is called an empirical equation, and smoothness is set by hand.
  • Image patch size = 16x16 pixels (first), 11x11 (second)
    Selected based on validation loss (Fig. S1); this determines what local structure is assumed to control the spectrum.
  • Im2spec latent dimension = 3
    Stated in Section III.A; no ablation or justification beyond design choice.
  • Initial labeled points = 245 (20%) for first; 30 for second
    Initial dataset size and seed sampling strategy affect the learning curves and are chosen by the authors.
  • Ensemble retraining trigger probability = 10% of iterations
    Probabilistic trigger is a tuning choice; no systematic study of its effect is presented.
assumptions (4)
  • domain assumption A fixed local image patch contains enough structural information to predict the measured spectrum (Im2spec) or to be predicted from it (Spec2im).
    The method assumes a local, learnable structure-property relationship; patch size is chosen by validation loss, and the paper notes that high-error regions may contain only noise.
  • ad hoc to paper The auxiliary model's predicted spectral mismatch error is a reliable signal for which unmeasured points will most improve the Im2spec/Spec2im model.
    The acquisition samples high predicted error; the paper acknowledges this can fail in plateaus, non-stationary errors, and noise-dominated points.
  • ad hoc to paper The empirical acquisition functions correctly trade exploration and exploitation.
    No derivation is given for the formulas; parameters beta and lambda are hand-set, and the paper studies extremes rather than optimizing them.
  • domain assumption Errors normalized by the running mean are sufficiently stationary for the error predictor to track.
    Mentioned as a fix for non-stationarity; even with this modification, training required a large learning rate and multiple epochs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Curiosity Driven Exploration to Optimize Structure-Property Learning in Microscopy." pith.science (2026). https://pith.science/paper/5DNANGDF

@misc{pith2026250420011,
  author       = {Pith},
  title        = {Pith review of: Curiosity Driven Exploration to Optimize Structure-Property Learning in Microscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5DNANGDF}},
  note         = {Machine review of arXiv:2504.20011}
}
read the original abstract

Rapidly determining structure-property correlations in materials is an important challenge in better understanding fundamental mechanisms and greatly assists in materials design. In microscopy, imaging data provides a direct measurement of the local structure, while spectroscopic measurements provide relevant functional property information. Deep kernel active learning approaches have been utilized to rapidly map local structure to functional properties in microscopy experiments, but are computationally expensive for multi-dimensional and correlated output spaces. Here, we present an alternative lightweight curiosity algorithm which actively samples regions with unexplored structure-property relations, utilizing a deep-learning based surrogate model for error prediction. We show that the algorithm outperforms random sampling for predicting properties from structures, and provides a convenient tool for efficient mapping of structure-property relationships in materials science.

Figures

Figures reproduced from arXiv: 2504.20011 by the authors.

Figure 1
Figure 1. FIG. 1: Description of the spectral search method on pre-acquired dataset. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Results of the error prediction model in active learning on the pre-acquired BEPS data. Panel [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Latent space representations for the active learning-based sampling. Panel [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4: Diagram of Curiosity Algorithm implementation with Im2spec used in conjunction with the [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Trial of Im2spec Curiosity Algorithm on pre-acquired PFM data: [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Minimum loss achieved by Im2spec with [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 8
Figure 8. Figure 8: FIG. 8: Trial of Curiosity Algorithm real-time on PFM microscope for Spec2im and Im2spec. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 24 canonical work pages

  1. [1]

    Segment anything for microscopy

    Anwai Archit, Luca Freckmann, Sushmita Nair, Nabeel Khalid, Paul Hilt, Vikas Rajashekar, Marei Freitag, Carolin Teuber, Genevieve Buckley, Sebastian von Haaren, et al. Segment anything for microscopy. Nature Methods , pages 1--13, 2025

    work page 2025

  2. [2]

    Scanning probe microscopy

    Ke Bian, Christoph Gerber, Andreas J Heinrich, Daniel J M \"u ller, Simon Scheuring, and Ying Jiang. Scanning probe microscopy. Nature Reviews Methods Primers , 1(1):36, 2021

    work page 2021

  3. [3]

    Exploration by random network distillation

    Yuri Burda, Harrison Edwards, Amos Storkey, and Oleg Klimov. Exploration by random network distillation. arXiv preprint arXiv:1810.12894 , 2018

    arXiv 2018

  4. [4]

    High-speed mapping of surface charge dynamics using sparse scanning kelvin probe force microscopy

    Marti Checa, Addis S Fuhr, Changhyo Sun, Rama Vasudevan, Maxim Ziatdinov, Ilia Ivanov, Seok Joon Yun, Kai Xiao, Alp Sehirlioglu, Yunseok Kim, et al. High-speed mapping of surface charge dynamics using sparse scanning kelvin probe force microscopy. Nature Communications , 14(1):7196, 2023

    work page 2023

  5. [5]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning , pages 1050--1059. PMLR, 2016

    work page 2016

  6. [6]

    A supervised learning approach for dynamic sampling

    GM Dilshan Godaliyadda, Dong Hye Ye, Michael D Uchic, Michael A Groeber, Gregery T Buzzard, and Charles A Bouman. A supervised learning approach for dynamic sampling. Electronic Imaging , 28:1--8, 2016

    work page 2016

  7. [7]

    Spiral high-speed scanning tunneling microscopy: Tracking atomic diffusion on the millisecond timescale

    Leonard Gura, Zechao Yang, Matthias Brinker, Florian Kala , William Kirstaedter, Patrik Marschalik, Heinz Junkes, Markus Heyde, and H-J Freund. Spiral high-speed scanning tunneling microscopy: Tracking atomic diffusion on the millisecond timescale. Applied Physics Letters , 119(25), 2021

    work page 2021

  8. [8]

    a se, Lo \

    Florian H \"a se, Lo \" c M Roch, and Al \'a n Aspuru-Guzik. Next-generation experimentation with self-driving laboratories. Trends in Chemistry , 1(3):282--291, 2019

    work page 2019

Show all 27 references
  1. [9]

    Atlas: a brain for self-driving laboratories

    Riley J Hickman, Malcolm Sim, Sergio Pablo-Garc \' a, Gary Tom, Ivan Woolhouse, Han Hao, Zeqing Bao, Pauric Bannigan, Christine Allen, Matteo Aldeghi, et al. Atlas: a brain for self-driving laboratories. Digital Discovery , 2025

    2025

  2. [10]

    Toward decoding the relationship between domain structure and functionality in ferroelectrics via hidden latent variables

    Sergei V Kalinin, Kyle Kelley, Rama K Vasudevan, and Maxim Ziatdinov. Toward decoding the relationship between domain structure and functionality in ferroelectrics via hidden latent variables. ACS Applied Materials & Interfaces , 13(1):1693--1703, 2021

    2021

  3. [11]

    Automated and autonomous experiments in electron and scanning probe microscopy

    Sergei V Kalinin, Maxim Ziatdinov, Jacob Hinkle, Stephen Jesse, Ayana Ghosh, Kyle P Kelley, Andrew R Lupini, Bobby G Sumpter, and Rama K Vasudevan. Automated and autonomous experiments in electron and scanning probe microscopy. ACS nano , 15(8):12604--12627, 2021

    2021

  4. [12]

    Demonstration of an ai-driven workflow for autonomous high-resolution scanning microscopy

    Saugat Kandel, Tao Zhou, Anakha V Babu, Zichao Di, Xinxin Li, Xuedan Ma, Martin Holt, Antonino Miceli, Charudatta Phatak, and Mathew J Cherukara. Demonstration of an ai-driven workflow for autonomous high-resolution scanning microscopy. Nature Communications , 14(1):5501, 2023

    2023

  5. [13]

    Artificial-intelligence-driven scanning probe microscopy

    Alexander Krull, Peter Hirsch, Carsten Rother, Augustin Schiffrin, and C Krull. Artificial-intelligence-driven scanning probe microscopy. Communications Physics , 3(1):54, 2020

    2020

  6. [14]

    Exploring physics of ferroelectric domain walls in real time: deep learning enabled scanning probe microscopy

    Yongtao Liu, Kyle P Kelley, Hiroshi Funakubo, Sergei V Kalinin, and Maxim Ziatdinov. Exploring physics of ferroelectric domain walls in real time: deep learning enabled scanning probe microscopy. Advanced Science , 9(31):2203957, 2022

    2022

  7. [15]

    Experimental discovery of structure--property relationships in ferroelectric materials via active learning

    Yongtao Liu, Kyle P Kelley, Rama K Vasudevan, Hiroshi Funakubo, Maxim A Ziatdinov, and Sergei V Kalinin. Experimental discovery of structure--property relationships in ferroelectric materials via active learning. Nature Machine Intelligence , 4(4):341--350, 2022

    2022

  8. [16]

    Aecroscopy: a software--hardware framework empowering microscopy toward automated and autonomous experimentation

    Yongtao Liu, Kevin Roccapriore, Marti Checa, Sai Mani Valleti, Jan-Chi Yang, Stephen Jesse, and Rama K Vasudevan. Aecroscopy: a software--hardware framework empowering microscopy toward automated and autonomous experimentation. Small Methods , 8(10):2301740, 2024

    2024

  9. [17]

    Exploring the relationship of microstructure and conductivity in metal halide perovskites via active learning-driven automated scanning probe microscopy

    Yongtao Liu, Jonghee Yang, Rama K Vasudevan, Kyle P Kelley, Maxim Ziatdinov, Sergei V Kalinin, and Mahshid Ahmadi. Exploring the relationship of microstructure and conductivity in metal halide perovskites via active learning-driven automated scanning probe microscopy. The Jour...

    2023

  10. [18]

    Deep learning for cellular image analysis

    Erick Moen, Dylan Bannon, Takamasa Kudo, William Graf, Markus Covert, and David Van Valen. Deep learning for cellular image analysis. Nature methods , 16(12):1233--1246, 2019

    2019

  11. [19]

    Multiscale structure-property discovery via active learning in scanning tunneling microscopy

    Ganesh Narasimha, Dejia Kong, Paras Regmi, Rongying Jin, Zheng Gai, Rama Vasudevan, and Maxim Ziatdinov. Multiscale structure-property discovery via active learning in scanning tunneling microscopy. arXiv preprint arXiv:2404.07074 , 2024

    2024 arXiv

  12. [20]

    A kriging-based approach to autonomous experimentation with applications to x-ray scattering

    Marcus M Noack, Kevin G Yager, Masafumi Fukuto, Gregory S Doerk, Ruipeng Li, and James A Sethian. A kriging-based approach to autonomous experimentation with applications to x-ray scattering. Scientific reports , 9(1):11809, 2019

    2019

  13. [21]

    Curiosity-driven exploration by self-supervised prediction

    Deepak Pathak, Pulkit Agrawal, Alexei A Efros, and Trevor Darrell. Curiosity-driven exploration by self-supervised prediction. In International conference on machine learning , pages 2778--2787. PMLR, 2017

    2017

  14. [22]

    Physics discovery in nanoplasmonic systems via autonomous experiments in scanning transmission electron microscopy

    Kevin M Roccapriore, Sergei V Kalinin, and Maxim Ziatdinov. Physics discovery in nanoplasmonic systems via autonomous experiments in scanning transmission electron microscopy. Advanced Science , 9(36):2203422, 2022

    2022

  15. [23]

    Bayesian conavigation: Dynamic designing of the material digital twins via active learning

    Boris N Slautin, Yongtao Liu, Hiroshi Funakubo, Rama K Vasudevan, Maxim Ziatdinov, and Sergei V Kalinin. Bayesian conavigation: Dynamic designing of the material digital twins via active learning. ACS nano , 18(36):24898--24908, 2024

    2024

  16. [24]

    Introduction to multi-armed bandits

    Aleksandrs Slivkins. Introduction to multi-armed bandits. arXiv , 2019

    2019

  17. [25]

    Curiosity in exploring chemical spaces: intrinsic rewards for molecular reinforcement learning

    Luca A Thiede, Mario Krenn, AkshatKumar Nigam, and Al \'a n Aspuru-Guzik. Curiosity in exploring chemical spaces: intrinsic rewards for molecular reinforcement learning. Machine Learning: Science and Technology , 3(3):035008, 2022

    2022

  18. [26]

    Schmid, Sterling G

    Gary Tom, Stefan P. Schmid, Sterling G. Baird, Yang Cao, Kourosh Darvish, Han Hao, Stanley Lo, Sergio Pablo-García, Ella M. Rajaonson, Marta Skreta, Naruki Yoshikawa, Samantha Corapi, Gun Deniz Akkoc, Felix Strieth-Kalthoff, Martin Seifrid, and Alán Aspuru-Guzik. Self-driving ...

    2024

  19. [27]

    Episodic multi-agent reinforcement learning with curiosity-driven exploration

    Lulu Zheng, Jiarui Chen, Jianhao Wang, Jiamin He, Yujing Hu, Yingfeng Chen, Changjie Fan, Yang Gao, and Chongjie Zhang. Episodic multi-agent reinforcement learning with curiosity-driven exploration. Advances in Neural Information Processing Systems , 34:3757--3769, 2021

    2021

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.