Pith. sign in

REVIEW 1 major objections 1 minor 37 references

MEDUSA: A Multimodal Deep Fusion Multi-Stage Training Framework for Speech Emotion Recognition in Naturalistic Conditions

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

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

Pith's one-line read A four-stage multimodal training pipeline—deep transformer fusion, soft annotation targets, balanced sampling, and a meta-classifier soup—was the top system on the categorical emotion recognition task of the challenge.

desk verdict The first-place challenge result is real and the system is a sensible engineering stack, but the ablation claims are overstated and internally inconsistent with the paper's own Table 2. read the letter →

arxiv 2506.09556 v2 pith:PNAANOF3 submitted 2025-06-11 cs.CL

Georgios Chatzichristodoulou , Despoina Kosmopoulou , Antonios Kritikos , Anastasia Poulopoulou , Efthymios Georgiou , Athanasios Katsamanis , Vassilis Katsouros , Alexandros Potamianos This is my paper · ORCID
classification cs.CL
keywords speechemotionrecognitionmultimodaldeepfusionsofttargetsclassimbalancemeta-classifiermodelsoupself-supervisedrepresentationsmulti-stagetraining
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

This paper claims that the hardest parts of speech emotion recognition in naturalistic audio—class imbalance and disagreement among human annotators—can be handled by a four-stage training recipe rather than by any single architectural innovation. The recipe combines deep cross-modal transformer fusion (DeepSER), soft classification targets built from full annotator vote vectors, balanced-subset fine-tuning with Manifold MixUp and multitask attribute regression, and a learned meta-classifier whose weights are averaged across seeds. Trained this way, an ensemble of 14 modality configurations ranked first in the challenge's categorical emotion recognition task, with macro-F1 of 0.4316, micro-F1 of 0.4319, and accuracy of 0.4319 on the held-out test set. A sympathetic reader would care because the result is a transferable template: each component is model-agnostic, so the recipe could be applied wherever emotional labels are subjective and skewed.

What carries the argument

Two objects carry the argument. The first is DeepSER, a deep fusion encoder: each modality is processed by a two-layer transformer encoder whose intermediate hidden states are concatenated and fed hierarchically into fusion transformer layers, so the final pooled representation combines unimodal and cross-modal features at multiple depths; its two heads predict the emotion class and the three emotional attributes, and Manifold MixUp interpolates latent features and labels before the classification head. The second is the four-stage training schedule: (1) standard training on the full dataset with softened annotator-vote targets and class-weighted loss; (2) continued training on rebalanced subsets with uniform class weights; (3) a linear meta-classifier trained on the ensemble's posterior vectors under the same balanced sampling; (4) a model-soup average of meta-classifier weights across random seeds. The machinery works by converting two data problems—annotator disagreement and skewed class frequencies—into training signals: soft targets preserve disagreement as information, and balanced stages force both base models and the meta-classifier to see minority emotions.

What would settle it

Train the same 14-model ensemble with one-hot labels instead of soft targets and evaluate on the official held-out test set; the paper's stated 8.23 percent relative macro-F1 drop from removing soft targets predicts a clear drop, so a much smaller or reversed gap would undercut the central role claimed for soft targets.

Watch

Extended reading notes

Core claim

The paper's central claim is that MEDUSA—four training stages wrapped around 14 DeepSER models—is the top system on this challenge's categorical emotion recognition task, and that each of its components earns its place. DeepSER is presented as a generalization of deep hierarchical fusion: unimodal transformer encoders produce intermediate representations that are fed into successive fusion transformer layers, so later fusion stages see both fresh unimodal features and earlier fusion outputs; the pooled concatenation then feeds a classification head and an arousal/valence/dominance regression head. Stage 1 trains each model on the full dataset with soft targets and class-weighted cross-entropy; Stage 2 continues on balanced subsets; Stage 3 trains a linear meta-classifier on ensemble posteriors; Stage 4 weight-averages meta-classifier runs. The ablations rank soft targets as the biggest single contributor (8.23 percent relative macro-F1 loss when removed), followed by deep fusion (6.31 percent), Manifold MixUp (2.93 percent), multitask learning (2.18 percent), and Stage 2 balanced training (1.20 percent). On the internal test split the full ensemble reaches macro-F1 of 0.472, a 7.23 percent relative gain over the best single DeepSER model, and on the official held-out test set the submitted system reports macro-F1 of 0.4316.

Load-bearing premise

The internal 90-10 train-test split used to choose the 14 models, loss weights, and soup variant is representative of the official held-out test set, so choices made on it transfer to the leaderboard.

Editorial extensions

If this is right

  • Using full annotator vote vectors instead of one-hot labels is the highest-leverage decision at the model level: the paper's ablation ties an 8.23 percent relative macro-F1 drop to removing it, so any naturalistic speech emotion recognition system with multiple annotators should preserve disagreement as a target distribution.
  • Deep fusion through stacked cross-modal transformer layers beats late fusion by 6.31 percent relative, implying that where to combine modalities matters as much as which modalities to combine.
  • A trainable linear meta-classifier over ensemble posteriors beats hard and soft voting, and weight-averaging its runs adds further gain as the ensemble grows to 14 models, so aggregation should be learned rather than fixed.
  • The nearly equal macro and micro F1 on the held-out set (0.4316 versus 0.4319) is the expected signature of successful imbalance handling; the same training recipe should transfer to other skewed affective tasks.
  • The meta-classifier soup outperforms a single meta-classifier only when the ensemble is large enough, so the benefit of weight averaging depends on having sufficient model diversity to average over.

Reading between the lines

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

  • The component rankings come from a single in-house 90-10 split, so an obvious extension is to repeat model selection and ablation on several random splits or on the challenge's own development partitions; stable rankings across splits would make the recipe's contribution easier to separate from the chosen split.
  • Because the meta-classifier improves with ensemble diversity more than with individual model quality, an explicit diversity measure over the 14 posterior vectors could predict when adding another pretrained speech or text encoder helps and when it is redundant.
  • The system uses automatic speech recognition transcripts at both train and test time, so part of the linguistic gain may be tied to recognition quality; comparing against human transcripts would separate the contribution of linguistic content from robustness to recognition errors.
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

1 major / 1 minor

Summary. The paper proposes MEDUSA, a multimodal framework for categorical speech emotion recognition in the Interspeech 2025 SER Challenge. The framework comprises a four-stage training pipeline: Stage 1 trains DeepSER, a transformer-based deep fusion architecture over self-supervised acoustic and linguistic representations, on the full training set; Stage 2 continues training on balanced subsets; Stage 3 trains a linear meta-classifier over an ensemble of DeepSER models; and Stage 4 averages meta-classifier weights via model soup. The method also uses soft annotation targets, multitask learning for emotional attributes, and Manifold MixUp regularization. MEDUSA is reported to rank first on the official leaderboard, with Macro-F1 0.4316, Micro-F1 0.4319, and Accuracy 0.4319 on the held-out test set. The paper includes DeepSER-level and MEDUSA-level ablation studies intended to attribute performance to the proposed components.

Significance. If the leaderboard result is accepted, the paper demonstrates a practical, competitive recipe for SER under naturalistic conditions, combining several known techniques with a transformer-based deep fusion architecture and a learned ensemble aggregation. The external challenge result is a concrete, non-circular validation, and the public code release is a strength. The methodological value is somewhat reduced by the ablation analysis: the paper's central explanatory claim that every component contributes consistently is not supported by the reported numbers, and the absence of variance estimates or significance tests weakens the causal interpretation. Nevertheless, the top-line result is credible and the framework itself is of interest to the SER community.

major comments (1)
  1. [3.4 and 5.2] DeepSER's novelty is framed as an extension of DHF from LSTM to transformers and from two to N modalities, but the ablation only compares deep fusion against late fusion. There is no comparison against the original DHF-style hierarchical fusion or against a non-hierarchical transformer fusion baseline, so it is unclear whether the observed gains come from the deep hierarchical structure, the transformer backbone, or both. Adding such a baseline would make the architectural contribution more precise.
minor comments (1)
  1. [5.1] The sentence "This balance between macro and micro metrics" is somewhat overstated because Macro-F1, Micro-F1, and Accuracy are 0.4316, 0.4319, and 0.4319, respectively; the near-identity is informative for class balance but the wording could be more measured.

Circularity Check

0 steps flagged · score 1.0 of 10

No material circularity: the top-line result is an externally judged leaderboard score, and ablations are standard empirical comparisons under a fixed internal protocol.

full rationale

MEDUSA's headline claim is validated by an external leaderboard (Section 5.1: Macro-F1 of 0.4316, Micro-F1 of 0.4319, Accuracy of 0.4319), which cannot be circular because the held-out test set was not used in any fitting or model-selection step. The internal 90-10 split (Section 4) is used only to rank configurations and ablations; selecting ensemble size and soup variant on internal F1 (Table 3) is standard model selection, not a prediction forced by a fit. The ablation matrix (Table 2) is an empirical comparison under a fixed protocol; the text's assertion that every component consistently improves F1 is internally inconsistent with several rows (e.g., 'w/o multitask' F1 exceeds DeepSER for Models 2 and 9; 'w/o Stage 2' F1 exceeds DeepSER for Model 13), but that is an evidence-quality problem, not circularity. The only self-citations are DHF [10] and PowMix [17], co-authored by present authors; they are used as architectural inspiration and related-work conceptual support, not as proof of MEDUSA's results, and they are published peer-reviewed works with independent existence. No equation defines the target result as its own input; no fitted parameter is relabeled as a prediction; no uniqueness claim is imported from the authors' prior work. Therefore no circular step is identified.

Assumptions & free parameters 9 free parameters · 6 assumptions · 0 invented entities

The central result is an empirical system rather than a derivation. It rests on transferred SSL representations, annotation quality, and validation-split representativeness. All loss weights, MixUp settings, architecture sizes, ensemble size, and soup runs are chosen by hand or internal validation, so the specific numbers are not derived from first principles. No new physical or natural entities are postulated; DeepSER and the meta-classifier are software components with shipped code.

free parameters (9)
  • Loss weight λ1 (cross-entropy) = 1.5
    Chosen by hand in Section 3.8 to balance classification and attribute regression losses.
  • Loss weight λ2 (MSE) = 0.4
    Chosen by hand in Section 3.8 alongside λ1.
  • Manifold MixUp probability p = 0.3
    Set in Section 4 as the probability of applying MixUp before the classification head.
  • Beta distribution α for MixUp interpolation = not specified
    Section 3.7 invokes Beta(α, α) but never reports the value used.
  • Class-weight exponent in α_c = 0.5
    The formula α_c = (N / (|C| · f_c))^0.5 in Section 3.8 is an ad hoc choice for reweighting classes.
  • Number of DeepSER models in the ensemble = 14
    Selected in Section 5.3 based on diminishing internal-F1 returns when doubling from 7 to 14 models.
  • Number of runs in meta-classifier soup = 6
    Stage 4 in Figure 1 averages six different meta-classifier runs.
  • Hidden dimension and layer count of DeepSER encoders = 1024 dimensions, 2 layers
    Architecture choice stated in Section 4, not derived from analysis.
  • Learning rates and batch sizes = 1e-5 and 16 for Stages 1-2; 1e-3 and 128 for Stage 3
    Training hyperparameters listed in Section 4 and selected by validation.
assumptions (6)
  • domain assumption Pretrained self-supervised models (WavLM, HuBERT, Whisper-v3, RoBERTa, ModernBERT) provide transferable representations for SER.
    Invoked in Section 3.3; the framework inherits these representations without independently validating them for emotion recognition.
  • domain assumption MSP-Podcast annotations and challenge labels are reliable ground truth for emotion.
    Section 3.2 normalizes annotator decisions into soft targets, so the quality of the human annotations bounds the method.
  • domain assumption Manifold MixUp interpolation creates valid virtual training points for speech emotion recognition.
    Section 3.7 relies on Vicinal Risk Minimization and prior MixUp literature; the paper does not prove this holds for the speech/emotion domain.
  • ad hoc to paper The internal 90-10 split is representative of the official held-out test distribution.
    Section 4 uses this split for model selection and all ablations, and Section 5.3 chooses the final ensemble configuration from it.
  • domain assumption Deep hierarchical cross-modal fusion is beneficial for this task.
    Inherited from DHF [10] and applied in Algorithm 2 without a unimodal baseline comparison.
  • domain assumption Whisper ASR transcripts do not introduce a harmful distribution shift for emotion classification.
    Section 3.3 uses ASR consistently for train and test, but the effect of transcription errors is not measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MEDUSA: A Multimodal Deep Fusion Multi-Stage Training Framework for Speech Emotion Recognition in Naturalistic Conditions." pith.science (2026). https://pith.science/paper/PNAANOF3

@misc{pith2026250609556,
  author       = {Pith},
  title        = {Pith review of: MEDUSA: A Multimodal Deep Fusion Multi-Stage Training Framework for Speech Emotion Recognition in Naturalistic Conditions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNAANOF3}},
  note         = {Machine review of arXiv:2506.09556}
}
read the original abstract

SER is a challenging task due to the subjective nature of human emotions and their uneven representation under naturalistic conditions. We propose MEDUSA, a multimodal framework with a four-stage training pipeline, which effectively handles class imbalance and emotion ambiguity. The first two stages train an ensemble of classifiers that utilize DeepSER, a novel extension of a deep cross-modal transformer fusion mechanism from pretrained self-supervised acoustic and linguistic representations. Manifold MixUp is employed for further regularization. The last two stages optimize a trainable meta-classifier that combines the ensemble predictions. Our training approach incorporates human annotation scores as soft targets, coupled with balanced data sampling and multitask learning. MEDUSA ranked 1st in Task 1: Categorical Emotion Recognition in the Interspeech 2025: Speech Emotion Recognition in Naturalistic Conditions Challenge.

Figures

Figures reproduced from arXiv: 2506.09556 by the authors.

Figure 1
Figure 1. An overview of MEDUSA. Different modalities are combined on each model using DeepSER. In total, 14 differ￾ent such models are trained leveraging 2 or 3 modalities as their backbone. The training process employs standard train￾ing (Stage 1) on the full training dataset, balanced subset train￾ing (Stage 2) and a linear meta-classifier on top for ensembling (Stage 3). Six different runs of the meta-classifier are merge… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 33 canonical work pages

  1. [1]

    Categorical Emo- tion Recognition

    Introduction Speech Emotion Recognition (SER) is a fundamental affective computing problem that aims to automatically identify emo- tional states from human speech [1], with significant technolog- ical applications in healthcare [2], intelligent driving [3], and call centers [4]. Despite the advancement in deep learning (DL) methodologies [5, 6], SER face...

    work page 2025

  2. [2]

    MEDUSA: A Multimodal Deep Fusion Multi-Stage Training Framework for Speech Emotion Recognition in Naturalistic Conditions

    Related Work SER spans a wide range of setups, from acted emotions [11] to naturalistic [8] and in-the-wild scenarios [12]. Research ap- proaches are typically supervised and use raw speech signals, mel-spectrograms and extracted features [13]. Other learning approaches employ self-supervised learning (SSL), as shown by emotion2vec [5] and WavLLM [14]. Mu...

    work page Pith review arXiv 2025

  3. [3]

    Overview The novelty of our approach stems from three key directions: 1) architecture, 2) data utilization and 3) training recipe

    Method 3.1. Overview The novelty of our approach stems from three key directions: 1) architecture, 2) data utilization and 3) training recipe. The pro- posed architecture, DeepSER, is a novel extension of a deep fu- sion approach [10]. Different DeepSER instances are combined via a meta-classifier and form M EDUSA . For data utilization, MEDUSA leverages ...

    work page 2025

  4. [4]

    The Stage 3 meta-classifier uses batch size 128 and learning rate 1e-3 without M.MixUp

    Experimental Setup DeepSER models are trained with batch size 16 and learning rate 1e-5 during Stages 1 & 2, with M.MixUp’s p = 0.3. The Stage 3 meta-classifier uses batch size 128 and learning rate 1e-3 without M.MixUp. DeepSER utilizes transformer en- coders (see Algorithm 1) with two layers across both unimodal and fusion processing, with hidden dimens...

    work page

  5. [5]

    Experimental Results 5.1. Challenge Result MEDUSA ranked 1st on the Challenge’s leaderboard, demon- strating remarkably consistent performance across all metrics on the held-out test set: Macro-F1 of 0.4316, Micro-F1 of 0.4319, and Accuracy of 0.4319. This balance between macro and micro metrics suggests that M EDUSA effectively handles both majority and ...

    work page

  6. [6]

    Discussion In this paper, we present M EDUSA , a multimodal and multi-stage framework that achieved first place in the Inter- speech 2025 Challenge. The M EDUSA framework integrates DeepSER, a novel deep fusion architectural extension, with soft targets, mixing-based regularization, and multi-stage training to address annotation ambiguity and class imbala...

    work page 2025

  7. [7]

    Acknowledgments This work is supported by the European Union’s Horizon Eu- rope Research and Innovation Programme under Grant No 101061303, project PREMIERE (Performing arts in a new era: AI and VR tools for better understanding, preservation, enjoy- ment and accessibility). This work has been partially supported by project MIS 5154714 of the National Rec...

    work page

  8. [8]

    Emotion recognition in human- computer interaction,

    R. Cowie, E. Douglas-Cowie, N. Tsapatsoulis, G. V otsis, S. Kol- lias, W. Fellenz, and J. Taylor, “Emotion recognition in human- computer interaction,”IEEE Signal Processing Magazine, vol. 18, no. 1, pp. 32–80, 2001

    work page 2001

Show all 37 references
  1. [9]

    Make patient consultation warmer: A clinical application for speech emotion recognition,

    H.-C. Li, T. Pan, M.-H. Lee, and H.-W. Chiu, “Make patient consultation warmer: A clinical application for speech emotion recognition,” Applied Sciences, vol. 11, no. 11, 2021

    2021

  2. [10]

    Integration of driver be- havior into emotion recognition systems: A preliminary study on steering wheel and vehicle acceleration,

    S. Shafaei, T. Hacizade, and A. Knoll, “Integration of driver be- havior into emotion recognition systems: A preliminary study on steering wheel and vehicle acceleration,” in Computer Vision – ACCV 2018 Workshops, G. Carneiro and S. You, Eds. Cham: Springer International Publi...

    2018

  3. [11]

    Call redistribution for a call center based on speech emotion recognition,

    M. Bojani ´c, V . Deli´c, and A. Karpov, “Call redistribution for a call center based on speech emotion recognition,” Applied Sciences, vol. 10, no. 13, 2020

    2020

  4. [12]

    emotion2vec: Self-supervised pre-training for speech emotion representation,

    Z. Ma, Z. Zheng, J. Ye, J. Li, Z. Gao, S. Zhang, and X. Chen, “emotion2vec: Self-supervised pre-training for speech emotion representation,” in Findings of the Association for Computational Linguistics: ACL 2024 , L.-W. Ku, A. Martins, and V . Srikumar, Eds. Bangkok, Thailand:...

    2024

  5. [13]

    Adieu features? end-to-end speech emotion recognition using a deep convolutional recurrent network,

    G. Trigeorgis, F. Ringeval, R. Brueckner, E. Marchi, M. A. Nico- laou, B. Schuller, and S. Zafeiriou, “Adieu features? end-to-end speech emotion recognition using a deep convolutional recurrent network,” in 2016 IEEE International Conference on Acoustics, Speech and Signal Pro...

    2016

  6. [14]

    The interspeech 2025 challenge on speech emotion recognition in naturalistic conditions,

    A. Reddy Naini, L. Goncalves, A. N. Salman, P. Mote, I. R.¨Ulgen, T. Thebaud, L. Velazquez, L. P. Garcia, N. Dehak, B. Sisman, and C. Busso, “The interspeech 2025 challenge on speech emotion recognition in naturalistic conditions,” in Interspeech 2025, vol. Under submission, R...

    2025

  7. [15]

    Building naturalistic emotionally bal- anced speech corpus by retrieving emotional speech from existing podcast recordings,

    R. Lotfian and C. Busso, “Building naturalistic emotionally bal- anced speech corpus by retrieving emotional speech from existing podcast recordings,” IEEE Transactions on Affective Computing, vol. 10, no. 4, pp. 471–483, 2019

    2019

  8. [16]

    Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time,

    M. Wortsman, G. Ilharco, S. Y . Gadre, R. Roelofs, R. Gontijo- Lopes, A. S. Morcos, H. Namkoong, A. Farhadi, Y . Carmon, S. Kornblith, and L. Schmidt, “Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time,” 2022

    2022

  9. [17]

    Deep hier- archical fusion with application in sentiment analysis,

    E. Georgiou, C. Papaioannou, and A. Potamianos, “Deep hier- archical fusion with application in sentiment analysis,” in Inter- speech 2019, 2019, pp. 1646–1650

    2019

  10. [18]

    Iemocap: Interactive emotional dyadic motion capture database,

    C. Busso, M. Bulut, C.-C. Lee, A. Kazemzadeh, E. Mower, S. Kim, J. N. Chang, S. Lee, and S. S. Narayanan, “Iemocap: Interactive emotional dyadic motion capture database,”Language Resources and Evaluation, vol. 42, no. 4, pp. 335–359, 2008

    2008

  11. [19]

    Multimodal language analysis in the wild: Cmu-mosei dataset and interpretable dynamic fusion graph,

    A. Zadeh, P. P. Liang, S. Poria, E. Cambria, and L.-P. Morency, “Multimodal language analysis in the wild: Cmu-mosei dataset and interpretable dynamic fusion graph,” in Annual Meeting of the Association for Computational Linguistics , 2018. [Online]. Available: https://api.sem...

    2018

  12. [20]

    A survey of speech emotion recognition in natural environment,

    M. Shah Fahad, A. Ranjan, J. Yadav, and A. Deepak, “A survey of speech emotion recognition in natural environment,” Digital Signal Processing, vol. 110, p. 102951, 2021

    2021

  13. [21]

    WavLLM: Towards robust and adaptive speech large language model,

    S. Hu, L. Zhou, S. Liu, S. Chen, L. Meng, H. Hao, J. Pan, X. Liu, J. Li, S. Sivasankaran, L. Liu, and F. Wei, “WavLLM: Towards robust and adaptive speech large language model,” in Findings of the Association for Computational Linguistics: EMNLP 2024 , Y . Al-Onaizan, M. Bansal...

    2024

  14. [22]

    Soft-target training with ambiguous emotional ut- terances for dnn-based speech emotion classification,

    A. Ando, S. Kobashikawa, H. Kamiyama, R. Masumura, Y . Ijima, and Y . Aono, “Soft-target training with ambiguous emotional ut- terances for dnn-based speech emotion classification,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018...

    2018

  15. [23]

    Manifold mixup: Better represen- tations by interpolating hidden states,

    V . Verma, A. Lamb, C. Beckham, A. Najafi, I. Mitliagkas, D. Lopez-Paz, and Y . Bengio, “Manifold mixup: Better represen- tations by interpolating hidden states,” inInternational conference on machine learning. PMLR, 2019, pp. 6438–6447

    2019

  16. [24]

    Powmix: A versa- tile regularizer for multimodal sentiment analysis,

    E. Georgiou, Y . Avrithis, and A. Potamianos, “Powmix: A versa- tile regularizer for multimodal sentiment analysis,”IEEE Transac- tions on Audio, Speech, and Language Processing, vol. 32, 2024

    2024

  17. [25]

    Speech emotion recognition with multi-task learning,

    X. Cai, J. Yuan, R. Zheng, L. Huang, and K. Church, “Speech emotion recognition with multi-task learning,” in Interspeech 2021, 2021, pp. 4508–4512

    2021

  18. [26]

    1st place solution to odyssey emotion recognition challenge task1: Tackling class imbalance problem,

    M. Chen, H. Zhang, Y . Li, J. Luo, W. Wu, Z. Ma, P. Bell, C. Lai, J. Reiss, L. Wang et al., “1st place solution to odyssey emotion recognition challenge task1: Tackling class imbalance problem,” arXiv preprint arXiv:2405.20064, 2024

    2024 arXiv

  19. [27]

    Meta-classifiers easily im- prove commercial sentiment detection tools,

    M. Cieliebak, O. D ¨urr, and F. Uzdilli, “Meta-classifiers easily im- prove commercial sentiment detection tools,” in Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC‘14) , N. Calzolari, K. Choukri, T. Declerck, H. Loftsson, B. Maega...

    2014

  20. [28]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems , I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garn...

    2017

  21. [29]

    Wavlm: Large-scale self-supervised pre-training for full stack speech processing,

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y . Qian, Y . Qian, J. Wu, M. Zeng, X. Yu, and F. Wei, “Wavlm: Large-scale self-supervised pre-training for full stack speech processing,” IEEE Journal of Selecte...

    2022

  22. [30]

    Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdi- nov, and A. Mohamed, “Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,” 2021

    2021

  23. [31]

    Robust speech recognition via large-scale weak su- pervision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” 2022

    2022

  24. [32]

    Roberta: A robustly optimized bert pretraining approach,

    Y . Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V . Stoyanov, “Roberta: A robustly optimized bert pretraining approach,” 2019

    2019

  25. [33]

    Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference,

    B. Warner, A. Chaffin, B. Clavi ´e, O. Weller, O. Hallstr ¨om, S. Taghadouini, A. Gallagher, R. Biswas, F. Ladhak, T. Aarsen, N. Cooper, G. Adams, J. Howard, and I. Poli, “Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long cont...

    2024

  26. [34]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  27. [35]

    Vicinal risk minimization,

    O. Chapelle, J. Weston, L. Bottou, and V . Vapnik, “Vicinal risk minimization,” in Advances in Neural Information Processing Systems, T. Leen, T. Dietterich, and V . Tresp, Eds., vol. 13. MIT Press, 2000

    2000

  28. [36]

    Decoupled weight decay regulariza- tion,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regulariza- tion,” 2019

    2019

  29. [37]

    Py- torch: An imperative style, high-performance deep learning li- brary,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. K ¨opf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chil- amkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “Py- torch: An imperative style, h...

    2019

Pith tools

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