FG 2025 TrustFAA: the First Workshop on Towards Trustworthy Facial Affect Analysis: Advancing Insights of Fairness, Explainability, and Safety (TrustFAA)

Harold Soh; Hatice Gunes; Jiaee Cheong; Yang Liu

REVIEW 1 major objections 4 minor 26 references

FG 2025 TrustFAA: the First Workshop on Towards Trustworthy Facial Affect Analysis: Advancing Insights of Fairness, Explainability, and Safety (TrustFAA)

T0 review · 1 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a first dedicated workshop on trustworthy facial affect analysis, arguing that Emotion AI tools deployed in high-stakes settings need focused attention on fairness, explainability, uncertainty, and privacy.

desk verdict A workshop proposal, not a research paper—cleanly organized and timely, but with no scientific claim to evaluate. read the letter →

arxiv 2506.05095 v1 pith:JSCWZIOF submitted 2025-06-05 cs.CV

Jiaee Cheong , Yang Liu , Harold Soh , Hatice Gunes This is my paper

classification cs.CV

keywords trustworthyAIfacialaffectanalysisfairnessexplainabilitysafetyEmotionaffectivecomputingworkshop

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

TrustFAA is a workshop proposal, not a technical study: it argues that facial affect analysis powered by Emotion AI is being used in high-stakes settings such as healthcare, education, driving, and human-robot interaction, and that the fairness, explainability, and safety of these systems deserve dedicated research attention. The paper lays out a half-day hybrid workshop that would bring together researchers from affective computing, trustworthy machine learning, and human-robot interaction, with invited talks, an oral session, a poster session, and peer-reviewed papers published in the workshop proceedings. A sympathetic reader would take the central claim to be that this dedicated, cross-community venue is the right next step for the field, and that the proposed scope, covering bias, uncertainty, interpretability, privacy, metrics, and ethics, captures the open problems. The paper also states concrete organizational expectations, including roughly 10-15 submissions and 45-50 attendees.

What carries the argument

The mechanism is the workshop itself as a structured venue. A tentative half-day schedule (welcome, three invited talks, oral session, poster session, closing) and a deliberately broad scope list define what counts as a contribution, while a double-blind review process by an internal and external programme committee selects papers for the proceedings. The hybrid format and the optional ethical and adverse-impact statements are additional design choices intended to widen participation and align with the host conference's ethics emphasis. What carries the argument is the assumption that assembling these elements will produce the exchange and gap-identification the paper promises.

What would settle it

Compare the published FG 2025 workshop program with the proposal: if fewer than roughly ten accepted papers appear, or if the half-day schedule runs with only two invited talks or no poster session, the paper's estimate that 10-15 submissions and 45-50 attendees would support the planned program has not held.

Watch

Extended reading notes

Core claim

The central claim is that trustworthy facial affect analysis should be treated as a research area of its own, with a dedicated venue to advance it. The paper describes the first such workshop, TrustFAA, at FG 2025, and specifies its program: three invited talks, eight-minute oral presentations with two-minute Q&A, a poster session and discussion, and a scope list covering trustworthy ML methods, fairness and bias mitigation, robustness under real-world variability, user-centered explainability, privacy-preserving facial affect analysis, assessment and standardization metrics, and ethical and social impacts. It argues that Emotion AI-powered facial affect analysis tools are increasingly deployed in settings with direct consequences for people, while trustworthiness challenges such as interpretability, uncertainty, biases, and privacy are under-investigated in existing trustworthy machine learning literature. The paper's contribution is organizational: creating the venue and soliciting double-blind-reviewed papers that will appear in the FG workshop proceedings.

Load-bearing premise

The plan assumes that at least roughly 10-15 papers will be submitted and about 45-50 people will attend; if actual interest is much lower, the half-day schedule with three invited talks and an oral session cannot be filled as proposed.

Editorial extensions

If this is right

If the estimated 10-15 submissions materialize, the workshop will publish a first set of peer-reviewed papers on trustworthy facial affect analysis in the FG 2025 proceedings.
A successful TrustFAA would create a common meeting ground for affective computing, trustworthy ML, and human-robot interaction researchers around fairness, explainability, and safety.
The invited talks and discussions are expected to identify existing gaps and open challenges in trustworthy machine learning for facial affect analysis.
The workshop's scope explicitly includes assessment and standardization of metrics, so accepted work may seed benchmark or evaluation protocols for fairness, uncertainty, and privacy in facial affect analysis.
By encouraging ethics and privacy topics, the workshop supports the host conference's requirement that submissions include an Ethical Impact Statement.

Reading between the lines

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

Editorial inference: a recurring TrustFAA could become the de facto venue where benchmark-style evaluations of fairness and uncertainty in facial affect analysis are proposed and compared, because the scope lists standardization but the paper itself does not commit to launching any benchmark.
Editorial inference: the position-paper track (2-4 pages) is a low-cost channel for critiques and early ideas; if used, it may surface critical perspectives on Emotion AI that full technical papers rarely include.
Editorial inference: the hybrid format plus the reliance on community mailing lists means the workshop's actual reach depends heavily on how well the organizers' existing networks overlap with the host conference community; a mismatch would show up quickly in the submission count.
Editorial inference: a testable extension is to compare accepted TrustFAA papers with the broader FG 2025 program on the dimensions of fairness, uncertainty, and privacy; a large overlap would suggest the workshop is serving needs the main conference already covers, while little overlap would support the paper's gap argument.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick answer: this is not a research paper; it is a workshop proposal for FG 2025, and the reader is right to mark it unverdictable as a scientific contribution. The only real claims are organizational: that a half-day workshop on fairness, explainability, and safety in facial affect analysis will bring together researchers. That is probably true and is a fine thing to propose.

What it does well: the topic is timely and the organizers are credible. The scope list is thoughtful and covers fairness, explainability, robustness, privacy, evaluation metrics, and ethics. The reference list anchors the proposal in relevant prior work, including the AI Act and the organizers' own publications. For a workshop solicitation, this does its job.

The soft spots exist but they are all outside the science. The paper was posted on 5 June 2025, after the paper submission deadline (10 March) and even after camera-ready (21 April). So the document reads as a proposal that was not updated post-hoc. That is fine if the arXiv version is meant to be a record of the proposal, but it could mislead readers into thinking the workshop has not happened yet. The estimates of 10-15 submissions and 45-50 attendees are guesses with no supporting data, and the schedule has a small AM/PM typo. None of this matters for any scientific evaluation because there is no scientific evaluation to make.

Who is this for? People looking for the text of the TrustFAA call for papers, or program committee members who want the workshop details. It is not for researchers looking for new methods or results. It does not deserve peer review as a paper; a serious editor would desk-reject it as a non-research artifact. As a workshop proposal, though, it is a reasonable piece of organizational writing. I would not cite it in my own work, and I would not bring it to reading group, though the topic itself is worth discussing if someone found actual papers on trustworthy facial affect analysis.

Referee Report

1 major / 4 minor

Summary. This manuscript, arXiv:2506.05095, presents a proposal for a half-day workshop titled 'TrustFAA' at FG 2025, focused on trustworthy facial affect analysis. It outlines the workshop's motivation (fairness, explainability, safety, and related concerns for Emotion AI), the planned program (three invited talks, an oral session, and a poster session), the paper submission and review procedure (double-blind, regular or position papers), and short biographies of the four organizers. The submission deadlines are given as 10 March, 30 March, and 21 April 2025, and the estimated turnout is 10–15 submissions and 45–50 attendees. The paper contains no experimental data, technical methods, or testable hypotheses; its content is organizational and promotional.

Significance. If the workshop is realized as described, it may provide a useful venue for connecting trustworthy ML with facial affect analysis, and the organizers have relevant prior work in fairness and affective computing. As a scholarly artifact, however, the paper makes no falsifiable claims and offers no evidence for the projected participation numbers. Its value is primarily agenda-setting: identifying open problems (bias, uncertainty, privacy, standardization) and assembling a community. The absence of any technical content is unsurprising for a workshop proposal, but it limits the manuscript to a descriptive role; no correctness assessment can be performed beyond checking that the planned schedule is internally consistent.

major comments (1)

[II-B (and arXiv header)] The manuscript is timestamped 5 June 2025, which is after all deadlines listed in Section II-B (paper submission 10 March, notification 30 March, camera-ready 21 April) and after the described half-day workshop would have taken place at FG 2025. The text is written in the future tense ('will be advertised,' 'expected to garner'), yet as posted it cannot serve as a call for papers and it does not report on the actual event. The central organizational claim—that TrustFAA will bring together researchers—is left ambiguous between a forward-looking proposal and a retrospective account. Please state explicitly whether this is a pre-workshop proposal submitted for the record or a post-workshop report, and if the latter, provide actual submission and attendance numbers.

minor comments (4)

[II-B, II-C] The figures 'approximately 10-15 submissions' and 'around 45-50 attendees' are asserted without a derivation or reference; a sentence describing how these estimates were formed (e.g., from comparable workshops) would improve the proposal's credibility.
[II-B] The text 'using the format provided by FG2025: here' relies on a hyperlink that may be lost in print; please include the full URL.
[References] Several references are missing bibliographic detail: [3] gives only '2024', [4] has no year or venue, and [17] is an arXiv preprint without an identifier. These should be completed for archival consistency.
[Section III] The 'Relevant publications' listed under each organizer duplicate entries already present in the main reference list; cross-referencing them would reduce redundancy.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: this workshop proposal makes no scientific derivation whose conclusion is presupposed by its inputs.

full rationale

This arXiv item is a workshop proposal and call for papers for TrustFAA at FG 2025, not a research paper with a derivation chain. Its central claims are organizational: that a workshop on trustworthy facial affect analysis is needed, that it will bring together researchers, and that a half-day schedule with invited speakers, an oral session, a poster session, and an expected 10-15 submissions and 45-50 attendees will take place. None of these claims is derived from a model, an equation, or a fitted parameter, so there is no input-to-output reduction that could be circular. The organizers' own prior publications appear only in biographical sketches and reference lists, as background context for their expertise, and are not used as evidence for any technical result. Likewise, the cited references motivate the importance of trustworthiness but do not function as a load-bearing proof step. The observation that the arXiv posting date (5 June 2025) falls after the stated internal deadlines (10 March, 30 March, 21 April 2025) is an operational oddity about the document's timing, not a circularity, and it does not affect any scientific claim because no scientific claim is made. Therefore, under the stated criteria, there is no significant circularity; the appropriate finding is a score of 0 with no specific circular steps.

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

This is a workshop proposal with no mathematical model, fitting procedure, or new entities. There are no free parameters, axioms, or invented constructs to audit.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FG 2025 TrustFAA: the First Workshop on Towards Trustworthy Facial Affect Analysis: Advancing Insights of Fairness, Explainability, and Safety (TrustFAA)." pith.science (2026). https://pith.science/paper/JSCWZIOF

@misc{pith2026250605095,
  author       = {Pith},
  title        = {Pith review of: FG 2025 TrustFAA: the First Workshop on Towards Trustworthy Facial Affect Analysis: Advancing Insights of Fairness, Explainability, and Safety (TrustFAA)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JSCWZIOF}},
  note         = {Machine review of arXiv:2506.05095}
}

read the original abstract

With the increasing prevalence and deployment of Emotion AI-powered facial affect analysis (FAA) tools, concerns about the trustworthiness of these systems have become more prominent. This first workshop on "Towards Trustworthy Facial Affect Analysis: Advancing Insights of Fairness, Explainability, and Safety (TrustFAA)" aims to bring together researchers who are investigating different challenges in relation to trustworthiness-such as interpretability, uncertainty, biases, and privacy-across various facial affect analysis tasks, including macro/ micro-expression recognition, facial action unit detection, other corresponding applications such as pain and depression detection, as well as human-robot interaction and collaboration. In alignment with FG2025's emphasis on ethics, as demonstrated by the inclusion of an Ethical Impact Statement requirement for this year's submissions, this workshop supports FG2025's efforts by encouraging research, discussion and dialogue on trustworthy FAA.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

26 extracted references · 23 canonical work pages

[1]

Cambria, X

E. Cambria, X. Zhang, R. Mao, M. Chen, and K. Kwok. Senticnet 8: Fusing emotion ai and commonsense ai for interpretable, trustworthy, and explainable affective computing. In HCII, 2024

work page 2024
[2]

Cetinkaya, S

B. Cetinkaya, S. Kalkan, and E. Akbas. Ranked: Addressing imbalance and uncertainty in edge detection using ranking-based losses. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3239–3249, 2024

work page 2024
[3]

Cheong, A

J. Cheong, A. Bangar, S. Kalkan, and H. Gunes. U-fair: Uncertainty- based multimodal multitask learning for fairer depression detection. 2024

work page 2024
[4]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Fairrefuse: Referee-guided fusion for multimodal causal fairness in depression detection

work page
[5]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. The hitchhiker’s guide to bias and fairness in facial affective signal processing: Overview and techniques. IEEE Signal Processing Magazine , 38(6):39–49, 2021

work page 2021
[6]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Causal structure learning of bias for fair affect recognition. In WACV 2023, pages 340–349, Jan 2023

work page 2023
[7]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Counterfactual fairness for facial expression recognition. In ECCV 2022 Workshops , pages 245–261. Springer, 2023

work page 2022
[8]

Cheong, S

J. Cheong, S. Kuzucu, S. Kalkan, and H. Gunes. Towards gender fairness for mental health prediction. In IJCAI, pages 5932–5940, 2023

work page 2023

Show all 26 references

[9]

it’s not fair!

J. Cheong, M. Spitale, and H. Gunes. “it’s not fair!”–fairness for a small dataset of multi-modal dyadic mental well-being coaching. In 2023 11th International Conference on Affective Computing and Intelligent Interaction (ACII) , pages 1–8. IEEE, 2023

2023
[10]

Cheong, M

J. Cheong, M. Spitale, and H. Gunes. Small but fair! fairness for multimodal human-human and robot-human mental wellbeing coaching. arXiv preprint arXiv:2407.01562 , 2024

2024 arXiv
[11]

Churamani, O

N. Churamani, O. Kara, and H. Gunes. Domain-incremental continual learning for mitigating bias in facial expression and action unit recognition. IEEE Transactions on Affective Computing , 14(4):3191– 3206, 2022

2022
[12]

Corti ˜nas-Lorenzo and G

K. Corti ˜nas-Lorenzo and G. Lacey. Toward explainable affective computing: A review. IEEE Transactions on Neural Networks and Learning Systems, 2023

2023
[13]

Devillers and R

L. Devillers and R. Cowie. Ethical considerations on affective computing: an overview. Proceedings of the IEEE , 2023

2023
[14]

European Commission et al

S. European Commission et al. Ethics guidelines for trustworthy ai. Publications Office, 2019

2019
[15]

B. C. Kok and H. Soh. Trust in robots: Challenges and opportunities. Current Robotics Reports , 1(4):297–309, 2020

2020
[16]

Kuzucu, J

S. Kuzucu, J. Cheong, H. Gunes, and S. Kalkan. Uncertainty as a fairness measure. Journal of Artificial Intelligence Research , 81:307– 335, 2024

2024
[17]

A. M. H. Kwok, J. Cheong, S. Kalkan, and H. Gunes. Machine learning fairness for depression detection using eeg data. arXiv preprint arXiv:2501.18192, 2025

2025 arXiv
[18]

H. Liu, Y . Wang, W. Fan, X. Liu, Y . Li, S. Jain, Y . Liu, A. Jain, and J. Tang. Trustworthy ai: A computational perspective. ACM Transactions on Intelligent Systems and Technology, 14(1):1–59, 2022

2022
[19]

Y . Liu, X. Zhang, J. Kauttonen, and G. Zhao. Uncertain label correc- tion via auxiliary action unit graphs for facial expression recognition. In 2022 26th International Conference on Pattern Recognition (ICPR), pages 777–783. IEEE, 2022

2022
[20]

Y . Liu, X. Zhang, J. Kauttonen, and G. Zhao. Uncertain facial expres- sion recognition via multi-task assisted correction. IEEE Transactions on Multimedia, 2023

2023
[21]

Mejbri, F

N. Mejbri, F. Essalmi, M. Jemni, and B. A. Alyoubi. Trends in the use of affective computing in e-learning environments. Education and Information Technologies, pages 1–23, 2022

2022
[22]

Ngxande, J.-R

M. Ngxande, J.-R. Tapamo, and M. Burke. Bias remediation in driver drowsiness detection systems using generative adversarial networks. IEEE Access, 8:55592–55601, 2020

2020
[23]

Sangeetha, R

S. Sangeetha, R. R. Immanuel, S. K. Mathivanan, J. Cho, and S. V . Easwaramoorthy. An empirical analysis of multimodal affective computing approaches for advancing emotional intelligence in artificial intelligence for healthcare. IEEE Access, 2024

2024
[24]

C. Saxena. Ethical considerations in affective computing. In Affective Computing for Social Good: Enhancing Well-being, Empathy, and Equity, pages 241–251. Springer, 2024

2024
[25]

Spitale, R

M. Spitale, R. Stower, M. T. Parreira, E. Yadollahi, I. Leite, and H. Gunes. Hri wasn’t built in a day: A call to action for responsible hri research. In 2024 33rd ROMAN, pages 696–702. IEEE, 2024

2024
[26]

T. Xu, J. White, S. Kalkan, and H. Gunes. Investigating bias and fairness in facial expression recognition. In Computer Vision–ECCV 2020 Workshops: Glasgow, UK, August 23–28, 2020, Proceedings, Part VI 16, pages 506–523. Springer, 2020

2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Cambria, X

E. Cambria, X. Zhang, R. Mao, M. Chen, and K. Kwok. Senticnet 8: Fusing emotion ai and commonsense ai for interpretable, trustworthy, and explainable affective computing. In HCII, 2024

work page 2024

[2] [2]

Cetinkaya, S

B. Cetinkaya, S. Kalkan, and E. Akbas. Ranked: Addressing imbalance and uncertainty in edge detection using ranking-based losses. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3239–3249, 2024

work page 2024

[3] [3]

Cheong, A

J. Cheong, A. Bangar, S. Kalkan, and H. Gunes. U-fair: Uncertainty- based multimodal multitask learning for fairer depression detection. 2024

work page 2024

[4] [4]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Fairrefuse: Referee-guided fusion for multimodal causal fairness in depression detection

work page

[5] [5]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. The hitchhiker’s guide to bias and fairness in facial affective signal processing: Overview and techniques. IEEE Signal Processing Magazine , 38(6):39–49, 2021

work page 2021

[6] [6]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Causal structure learning of bias for fair affect recognition. In WACV 2023, pages 340–349, Jan 2023

work page 2023

[7] [7]

Cheong, S

J. Cheong, S. Kalkan, and H. Gunes. Counterfactual fairness for facial expression recognition. In ECCV 2022 Workshops , pages 245–261. Springer, 2023

work page 2022

[8] [8]

Cheong, S

J. Cheong, S. Kuzucu, S. Kalkan, and H. Gunes. Towards gender fairness for mental health prediction. In IJCAI, pages 5932–5940, 2023

work page 2023