pith. sign in

arxiv: 2504.10808 · v3 · submitted 2025-04-15 · 💻 cs.CV · cs.HC· cs.LG

Privacy-Preserving Empathy Detection in Video Interactions

Md Rakibul Hasan , Md Zakir Hossain , Aneesh Krishna , Shafin Rahman , Tom Gedeon This is my paper

Pith reviewed 2026-05-22 20:46 UTC · model grok-4.3

classification 💻 cs.CV cs.HCcs.LG
keywords empathy detectionprivacy preservationtabular foundation modelsvideo interactionshuman-robot interactioncross-subject evaluationdata minimizationbehavioral prediction
0
0 comments X

The pith

Summary statistics of visual features suffice for subject-generalizable empathy detection under strong privacy.

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

The paper shows that empathy can be detected from video interactions using only summary statistics of features like facial landmarks, action units and eye gaze, without raw video or full temporal sequences. This matters for applications where privacy rules block video sharing, as it tests whether useful prediction survives the strongest privacy constraints. The authors define three privacy levels and introduce TFMPathy, which applies tabular foundation models to aggregated data on a human-robot benchmark. They add a cross-subject evaluation protocol and report that fine-tuning raises accuracy from 0.590 to 0.730 while aggregation reduces subject-specific cues.

Core claim

Strong, subject-generalisable empathy detection is achievable at the strong-privacy level. TFMPathy, built with tabular foundation models TabPFN v2 and TabICL under in-context learning and fine-tuning, reaches accuracy 0.730 and AUC 0.669 on a public human-robot interaction benchmark under a new cross-subject protocol, outperforming baselines while summary statistics suppress demographic information.

What carries the argument

TFMPathy, the application of tabular foundation models to summary statistics of temporal visual features for privacy-constrained prediction.

If this is right

  • Empathy detection becomes feasible in settings that prohibit raw video sharing.
  • Cross-subject protocols provide a stricter test of generalisation than prior splits.
  • Feature aggregation aligns model training with data-minimisation rules by reducing identifiable cues.
  • Tabular foundation models offer a practical method for behavioural prediction when only aggregates are available.
  • The approach supports AI deployment under institutional or consent restrictions on video data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same summary-statistic pipeline could be applied to other behavioural signals such as engagement or stress in privacy-sensitive domains.
  • Collaborative research might standardise on aggregate feature releases rather than raw or landmark streams.
  • Testing on datasets with wider demographic range would check whether cue suppression holds beyond the current benchmark.
  • Combining the tabular models with other modalities could further improve utility without relaxing privacy constraints.

Load-bearing premise

The summary statistics of temporal visual features retain enough signal for empathy prediction across subjects after aggregation removes most individual detail.

What would settle it

A new cross-subject test on the same or similar benchmark where no model using only summary statistics exceeds chance-level accuracy would falsify the central claim.

Figures

Figures reproduced from arXiv: 2504.10808 by Aneesh Krishna, Md Rakibul Hasan, Md Zakir Hossain, Shafin Rahman, Tom Gedeon.

Figure 1
Figure 1. Figure 1: Overview of the task and motivation. We study the task of detecting [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our proposed TFMPathy pipeline for empathy detection. Temporal multimodal features (e.g., facial landmarks, eye gaze) are first extracted [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: In-context learning performance of TFMPathy across different proportions of in-context data (i.e., training data used in the single forward pass of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

Detecting empathy from video interactions has emerging applications, yet raw videos that could be used for training AI models are rarely available due to privacy and ethical constraints. Public benchmarks are consequently released only as pre-extracted features, creating a privacy-constrained learning regime whose privacy-utility trade-off is poorly characterised. We formalise three levels of privacy for video-based behavioural prediction -- no privacy (raw video), partial privacy (temporal visual features such as facial landmarks, action units and eye gaze) and strong privacy (summary statistics of those features) -- and ask whether strong, subject-generalisable empathy detection is achievable at the strong-privacy level. We propose TFMPathy, instantiated with two recent Tabular Foundation Models (TFMs) (TabPFN v2 and TabICL), under both in-context learning and fine-tuning paradigms. On a public human-robot interaction benchmark, TFMPathy achieves strong utility under strong privacy, outperforming established baselines by a substantial margin. To assess robustness and facilitate fair, safe deployment, we introduce a cross-subject evaluation protocol that was previously lacking in this benchmark. Under this protocol, TFM fine-tuning improves generalisation capacity substantially (accuracy: $0.590 \rightarrow 0.730$; AUC: $0.564 \rightarrow 0.669$). Aggregating temporal features into summary statistics also suppresses subject-specific and demographic cues, aligning TFMPathy with data-minimisation principles. TFMPathy, therefore, offers a practical route to building AI systems that depend on human-centred video when governance, consent or institutional policies restrict the sharing of raw video. Code will be released upon acceptance at https://github.com/hasan-rakibul/TFMPathy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript formalizes three privacy levels for video-based empathy detection (no privacy: raw video; partial: temporal visual features; strong: summary statistics of landmarks/AUs/gaze) and proposes TFMPathy, which applies recent tabular foundation models (TabPFN v2, TabICL) under in-context learning and fine-tuning. On a public human-robot interaction benchmark it reports that fine-tuning yields accuracy gains from 0.590 to 0.730 and AUC gains from 0.564 to 0.669 under a newly introduced cross-subject protocol, while the aggregation step suppresses subject-specific and demographic cues.

Significance. If the reported gains are robust, the work would demonstrate that strong-privacy summary statistics can support subject-generalizable empathy prediction, providing a practical route for affective computing under data-minimisation constraints. The cross-subject protocol and use of tabular foundation models are timely contributions; explicit code release would further strengthen reproducibility.

major comments (2)
  1. [strong-privacy definition and results paragraph] The central claim that summary statistics preserve enough cross-subject empathy signal (abstract, strong-privacy definition) is load-bearing for the utility results. The manuscript should include an ablation or variance decomposition showing how much empathy-related variance remains after aggregation versus subject-specific components; without it the 0.140 accuracy and 0.105 AUC lifts cannot be confidently attributed to retained temporal dynamics rather than benchmark idiosyncrasies.
  2. [evaluation and results sections] Baseline implementations, dataset subject count, statistical significance tests, and exact aggregation functions (means, variances, etc.) are not detailed enough to reproduce or interpret the 'substantial margin' over established baselines. These omissions directly affect whether the cross-subject protocol improvements are generalizable.
minor comments (2)
  1. [abstract and introduction] Define all acronyms (TFM, AUC, TFMPathy) on first use and ensure consistent notation for the three privacy levels throughout.
  2. [results] Clarify whether the reported numbers are from a single run or averaged over multiple seeds; add error bars or confidence intervals to the accuracy/AUC figures.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which highlights important aspects for strengthening the manuscript's claims and reproducibility. We respond to each major comment below.

read point-by-point responses

  1. Referee: [strong-privacy definition and results paragraph] The central claim that summary statistics preserve enough cross-subject empathy signal (abstract, strong-privacy definition) is load-bearing for the utility results. The manuscript should include an ablation or variance decomposition showing how much empathy-related variance remains after aggregation versus subject-specific components; without it the 0.140 accuracy and 0.105 AUC lifts cannot be confidently attributed to retained temporal dynamics rather than benchmark idiosyncrasies.

    Authors: We agree that stronger evidence is needed to attribute the gains specifically to retained empathy signals. The manuscript demonstrates suppression of subject-specific and demographic cues via aggregation, which correlates with the observed cross-subject improvements. However, a dedicated variance decomposition was not included. We will add an ablation comparing aggregated vs. non-aggregated features and a discussion of variance components based on the available benchmark data. This constitutes a partial revision, as a complete decomposition may require supplementary experiments beyond the current scope. revision: partial

  2. Referee: [evaluation and results sections] Baseline implementations, dataset subject count, statistical significance tests, and exact aggregation functions (means, variances, etc.) are not detailed enough to reproduce or interpret the 'substantial margin' over established baselines. These omissions directly affect whether the cross-subject protocol improvements are generalizable.

    Authors: We acknowledge these omissions limit reproducibility. In the revised version we will expand the evaluation and results sections to specify: baseline implementations in full detail, the exact subject count in the benchmark, statistical significance tests (e.g., McNemar's test or paired t-tests on the reported metrics), and the precise aggregation functions (means, variances, and any other statistics). The planned code release will provide the exact implementations for verification. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical results on public benchmark are independent of inputs

full rationale

The paper presents an empirical evaluation of TFMPathy on a public human-robot interaction benchmark, reporting performance gains under a cross-subject protocol (accuracy 0.590→0.730, AUC 0.564→0.669) after defining privacy levels and applying tabular foundation models to summary statistics. No equations, derivations, or self-citations reduce these results to quantities defined by the paper's own fitted parameters or inputs by construction. The formalization of strong-privacy aggregation and the choice of in-context learning/fine-tuning are external modeling decisions tested against held-out subject splits, leaving the central claim falsifiable on the benchmark rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that aggregated summary statistics retain sufficient empathy signal and on standard machine-learning assumptions about model generalization; no free parameters or invented entities beyond the method name are detailed in the abstract.

axioms (1)
  • domain assumption Summary statistics of temporal visual features preserve enough information for subject-generalizable empathy detection.
    Invoked when defining the strong-privacy level and claiming it supports generalizable performance.
invented entities (1)
  • TFMPathy no independent evidence
    purpose: Name for the proposed tabular-foundation-model pipeline for privacy-preserving empathy detection
    New method introduced in the paper; no independent evidence outside this work.

pith-pipeline@v0.9.0 · 5862 in / 1360 out tokens · 47046 ms · 2026-05-22T20:46:05.162485+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages

  1. [1]

    Modeling user empathy elicited by a robot storyteller,

    L. Mathur, M. Spitale, H. Xi, J. Li, and M. J. Matari ´c, “Modeling user empathy elicited by a robot storyteller,” in 2021 9th International Conference on Affective Computing and Intelligent Interaction (ACII) , 2021, pp. 1–8. DOI: 10.1109/ACII52823.2021.9597416

    work page doi:10.1109/acii52823.2021.9597416 2021

  2. [2]

    M. H. Davis, Empathy: A Social Psychological Approach , 1st. Rout- ledge, 1996. DOI: 10.4324/9780429493898

    work page doi:10.4324/9780429493898 1996

  3. [3]

    M. L. Hoffman, Empathy and Moral Development: Implications for Caring and Justice . Cambridge University Press, 2000. DOI: 10.1017/ CBO9780511805851

    work page 2000

  4. [4]

    Empathy training in health sciences: A systematic review,

    P. Bas-Sarmiento, M. Fern ´andez-Guti´errez, M. Baena-Ba ˜nos, A. Correro-Bermejo, P. S. Soler-Martins, and S. de la Torre-Moyano, “Empathy training in health sciences: A systematic review,” Nurse Education in Practice , vol. 44, p. 102 739, 2020. DOI: 10 . 1016 / j . nepr.2020.102739

    work page arXiv 2020

  5. [5]

    Empathy detection from text, audiovisual, audio or physiological signals: A systematic review of task formulations and machine learning methods,

    M. R. Hasan, M. Z. Hossain, S. Ghosh, A. Krishna, and T. Gedeon, “Empathy detection from text, audiovisual, audio or physiological signals: A systematic review of task formulations and machine learning methods,” IEEE Transactions on Affective Computing, pp. 1–20, 2025. DOI: 10.1109/TAFFC.2025.3590107

    work page doi:10.1109/taffc.2025.3590107 2025

  6. [6]

    Modeling empathy and distress in reaction to news stories,

    S. Buechel, A. Buffone, B. Slaff, L. Ungar, and J. Sedoc, “Modeling empathy and distress in reaction to news stories,” in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Pro- cessing, Brussels, Belgium: Association for Computational Linguistics, Oct. 2018, pp. 4758–4765. DOI: 10.18653/v1/D18-1507

    work page doi:10.18653/v1/d18-1507 2018

  7. [7]

    Modeling empathic similarity in personal narratives,

    J. Shen, M. Sap, P. Colon-Hernandez, H. Park, and C. Breazeal, “Modeling empathic similarity in personal narratives,” in Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, H. Bouamor, J. Pino, and K. Bali, Eds., Singapore: Association for Computational Linguistics, Dec. 2023, pp. 6237–6252. DOI: 10.18653/v1/2023.emnlp-main.383

    work page doi:10.18653/v1/2023.emnlp-main.383 2023

  8. [8]

    The omg-empathy dataset: Evaluating the impact of affective behavior in storytelling,

    P. Barros, N. Churamani, A. Lim, and S. Wermter, “The omg-empathy dataset: Evaluating the impact of affective behavior in storytelling,” in 2019 8th International Conference on Affective Computing and Intelligent Interaction (ACII) , IEEE, 2019, pp. 1–7. DOI: 10 . 1109 / ACII.2019.8925530

    work page arXiv 2019

  9. [9]

    It takes two to empathize: One to seek and one to provide,

    M. Hosseini and C. Caragea, “It takes two to empathize: One to seek and one to provide,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, 2021, pp. 13 018–13 026. DOI: 10.1609/ aaai.v35i14.17539

    work page 2021

  10. [10]

    A theory of narrative empathy,

    S. Keen, “A theory of narrative empathy,” Narrative, vol. 14, no. 3, pp. 207–236, 2006

    work page 2006

  11. [11]

    Socially assistive robots as storytellers that elicit empathy,

    M. Spitale, S. Okamoto, M. Gupta, H. Xi, and M. J. Matari ´c, “Socially assistive robots as storytellers that elicit empathy,” ACM Transactions on Human-Robot Interaction, vol. 11, no. 4, Sep. 2022. DOI: 10.1145/ 3538409. UNDER REVIEW 9

    work page 2022

  12. [12]

    Tabpfn: A transformer that solves small tabular classification problems in a second,

    N. Hollmann, S. M ¨uller, K. Eggensperger, and F. Hutter, “Tabpfn: A transformer that solves small tabular classification problems in a second,” in Proceedings of the International Conference on Learning Representations (ICLR’23), 2023

    work page 2023

  13. [13]

    Accurate predictions on small data with a tab- ular foundation model.Nature, 637(8045):319–326, 2025

    N. Hollmann, S. M ¨uller, L. Purucker, et al., “Accurate predictions on small data with a tabular foundation model,”Nature, vol. 637, no. 8045, pp. 319–326, 2025. DOI: 10.1038/s41586-024-08328-6

    work page doi:10.1038/s41586-024-08328-6 2025

  14. [14]

    Tabllm: Few-shot classification of tabular data with large language models,

    S. Hegselmann, A. Buendia, H. Lang, M. Agrawal, X. Jiang, and D. Sontag, “Tabllm: Few-shot classification of tabular data with large language models,” in Proceedings of the 26th International Conference on Artificial Intelligence and Statistics, F. Ruiz, J. Dy, and J.-W. van de Meent, Eds., ser. Proceedings of Machine Learning Research, vol. 206, PMLR, Ap...

    work page 2023

  15. [15]

    Large language models (llms) on tabular data: Prediction, generation, and understanding - A survey,

    X. Fang, W. Xu, F. A. Tan, et al., “Large language models (llms) on tabular data: Prediction, generation, and understanding - A survey,” Transactions on Machine Learning Research , vol. 2024, 2024

    work page 2024

  16. [16]

    Transformers can do bayesian inference,

    S. M ¨uller, N. Hollmann, S. Pineda-Arango, J. Grabocka, and F. Hutter, “Transformers can do bayesian inference,” in Proceedings of the Tenth International Conference on Learning Representations (ICLR’22), OpenReview.net, 2022

    work page 2022

  17. [17]

    J. Qu, D. Holzm ¨uller, G. Varoquaux, and M. L. Morvan, Tabicl: A tabular foundation model for in-context learning on large data , 2025

    work page 2025

  18. [18]

    An empirical study of gpt-3 for few-shot knowledge-based vqa,

    Z. Yang, Z. Gan, J. Wang, et al. , “An empirical study of gpt-3 for few-shot knowledge-based vqa,” Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 3, pp. 3081–3089, Jun. 2022. DOI: 10.1609/aaai.v36i3.20215

    work page doi:10.1609/aaai.v36i3.20215 2022

  19. [19]

    Emotion recognition toolkit (ertk): Standardising tools for emotion recognition research,

    A. Keesing, Y . S. Koh, V . Yogarajan, and M. Witbrock, “Emotion recognition toolkit (ertk): Standardising tools for emotion recognition research,” in Proceedings of the 31st ACM International Conference on Multimedia , New York, NY , USA: Association for Computing Machinery, 2023, pp. 9693–9696. DOI: 10.1145/3581783.3613459

    work page doi:10.1145/3581783.3613459 2023

  20. [20]

    Curtin OCAI at W ASSA 2023 empathy, emotion and personality shared task: Demographic-aware prediction using multiple transform- ers,

    M. R. Hasan, M. Z. Hossain, T. Gedeon, S. Soon, and S. Rahman, “Curtin OCAI at W ASSA 2023 empathy, emotion and personality shared task: Demographic-aware prediction using multiple transform- ers,” in Proceedings of the 13th Workshop on Computational Ap- proaches to Subjectivity, Sentiment, & Social Media Analysis , Toronto, Canada: Association for Comput...

    work page 2023

  21. [21]

    DOI: 10.18653/v1/2023.wassa-1.47

    work page doi:10.18653/v1/2023.wassa-1.47 2023

  22. [22]

    Identifying empathetic messages in online health communities,

    H. Khanpour, C. Caragea, and P. Biyani, “Identifying empathetic messages in online health communities,” in Proceedings of the Eighth International Joint Conference on Natural Language Processing (Vol- ume 2: Short Papers) , Taipei, Taiwan: Asian Federation of Natural Language Processing, Nov. 2017, pp. 246–251

    work page 2017

  23. [23]

    LeadEmpathy: An expert annotated German dataset of empathy in written leadership communication,

    D. Sedefoglu, A. C. Lahnala, J. Wagner, L. Flek, and S. Ohly, “LeadEmpathy: An expert annotated German dataset of empathy in written leadership communication,” in Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , N. Calzolari, M.-Y . Kan, V . Hoste, A. Lenci, S. Sakt...

    work page 2024

  24. [24]

    Empathizing with virtual agents: The effect of personification and general empathic tendencies,

    K. Kroes, I. Saccardi, and J. Masthoff, “Empathizing with virtual agents: The effect of personification and general empathic tendencies,” in 2022 IEEE International Conference on Artificial Intelligence and Virtual Reality (AIVR) , IEEE, 2022, pp. 73–81. DOI: 10 . 1109 / AIVR56993.2022.00017

    work page arXiv 2022

  25. [25]

    Multimodal learning for identifying opportunities for em- pathetic responses,

    L. Tavabi, K. Stefanov, S. Nasihati Gilani, D. Traum, and M. So- leymani, “Multimodal learning for identifying opportunities for em- pathetic responses,” in 2019 International Conference on Multimodal Interaction, 2019, pp. 95–104. DOI: 10.1145/3340555.3353750

    work page doi:10.1145/3340555.3353750 2019

  26. [26]

    A multimodal framework for automated teaching quality assessment of one-to-many online instruction videos,

    Y . Pan, J. Wu, R. Ju, et al., “A multimodal framework for automated teaching quality assessment of one-to-many online instruction videos,” in 2022 26th International Conference on Pattern Recognition (ICPR) , IEEE, 2022, pp. 1777–1783. DOI: 10.1109/icpr56361.2022.9956185

    work page doi:10.1109/icpr56361.2022.9956185 2022

  27. [27]

    A learning system to support social and empathy disor- ders diagnosis through affective avatars,

    R. Herv ´as, E. Johnson, C. G. L. de la Franca, J. Bravo, and T. Mond´ejar, “A learning system to support social and empathy disor- ders diagnosis through affective avatars,” in 2016 15th International Conference on Ubiquitous Computing and Communications and 2016 International Symposium on Cyberspace and Security (IUCC-CSS) , IEEE, 2016, pp. 93–100. DOI:...

    work page doi:10.1109/iucc-css.2016.021 2016

  28. [28]

    MEDIC: A multimodal empathy dataset in counseling,

    Z. Zhu, C. Li, J. Pan, et al., “MEDIC: A multimodal empathy dataset in counseling,” in Proceedings of the 31st ACM International Conference on Multimedia , New York, NY , USA: Association for Computing Machinery, 2023, pp. 6054–6062. DOI: 10.1145/3581783.3612346

    work page doi:10.1145/3581783.3612346 2023

  29. [29]

    EmpathicStories++: A multimodal dataset for empathy towards personal experiences,

    J. Shen, Y . Kim, M. Hulse, et al., “EmpathicStories++: A multimodal dataset for empathy towards personal experiences,” in Findings of the Association for Computational Linguistics: ACL 2024 , L.-W. Ku, A. Martins, and V . Srikumar, Eds., Bangkok, Thailand: Association for Computational Linguistics, Aug. 2024, pp. 4525–4536. DOI: 10.18653/ v1/2024.finding...

    work page 2024

  30. [30]

    M. R. Hasan, Y . Yao, M. Z. Hossain, et al., Labels generated by large language models help measure people’s empathy in vitro , 2025

    work page 2025

  31. [31]

    RU at W ASSA 2024 shared task: Task-aligned prompt for predicting empathy and distress,

    H. Kong and S. Moon, “RU at W ASSA 2024 shared task: Task-aligned prompt for predicting empathy and distress,” inProceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, O. De Clercq, V . Barriere, J. Barnes, R. Klinger, J. Sedoc, and S. Tafreshi, Eds., Bangkok, Thailand: Association for Computation...

    work page 2024

  32. [32]

    LLM- GEm: Large language model-guided prediction of people’s empathy levels towards newspaper article,

    M. R. Hasan, M. Z. Hossain, T. Gedeon, and S. Rahman, “LLM- GEm: Large language model-guided prediction of people’s empathy levels towards newspaper article,” in Findings of the Association for Computational Linguistics: EACL 2024 , Y . Graham and M. Purver, Eds., St. Julian’s, Malta: Association for Computational Linguistics, Mar. 2024, pp. 2215–2231

    work page 2024

  33. [33]

    HIT-SCIR at W ASSA 2023: Empathy and emotion analysis at the utterance-level and the essay-level,

    X. Lu, Z. Li, Y . Tong, Y . Zhao, and B. Qin, “HIT-SCIR at W ASSA 2023: Empathy and emotion analysis at the utterance-level and the essay-level,” in Proceedings of the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis , Toronto, Canada: Association for Computational Linguistics, Jul. 2023, pp. 574–580. DOI: 10.1...

    work page doi:10.18653/v1/2023.wassa-1.54 2023

  34. [34]

    Frick and M

    R. Frick and M. Steinebach, “Fraunhofer SIT at W ASSA 2024 empathy and personality shared task: Use of sentiment transformers and data augmentation with fuzzy labels to predict emotional reactions in conversations and essays,” in Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, O. De Clercq,...

    work page doi:10.18653/v1/2024.wassa- 2024

  35. [35]

    van Rijn, Bernd Bischl, and Luis Torgo

    J. Vanschoren, J. N. van Rijn, B. Bischl, and L. Torgo, “Openml: Networked science in machine learning,” SIGKDD Explor. Newsl. , vol. 15, no. 2, pp. 49–60, Jun. 2014. DOI: 10.1145/2641190.2641198

    work page doi:10.1145/2641190.2641198 2014

  36. [36]

    OpenFace: A general-purpose face recognition library with mobile applications,

    B. Amos, B. Ludwiczuk, M. Satyanarayanan, et al. , “OpenFace: A general-purpose face recognition library with mobile applications,” CMU School of Computer Science , vol. 6, no. 2, p. 20, 2016

    work page 2016

  37. [37]

    Optuna: A next-generation hyperparameter optimization framework,

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” in Proceed- ings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , 2019, pp. 2623–2631. DOI: 10 . 1145 / 3292500.3330701

    work page arXiv 2019

  38. [38]

    Defazio, X

    A. Defazio, X. Yang, H. Mehta, K. Mishchenko, A. Khaled, and A. Cutkosky, The road less scheduled , 2024

    work page 2024