pith. sign in

arxiv: 1907.04294 · v1 · pith:UDK4DVWDnew · submitted 2019-07-09 · 💻 cs.IR · cs.SD· eess.AS

An Attention Mechanism for Musical Instrument Recognition

Siddharth Gururani , Mohit Sharma , Alexander Lerch This is my paper

Pith reviewed 2026-05-24 23:59 UTC · model grok-4.3

classification 💻 cs.IR cs.SDeess.AS
keywords attention mechanismmusical instrument recognitionweakly labeled datamulti-label classificationpolyphonic audioOpenMIC datasetinterpretability in audio models
0
0 comments X

The pith

An attention mechanism improves accuracy across all 20 instruments when recognizing multiple musical instruments from weakly labeled audio clips.

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

The paper tests whether adding an attention layer to a neural network can improve multi-label instrument recognition when training data only indicates which instruments appear in an entire clip rather than their exact timing. It evaluates the approach on the OpenMIC dataset against a random forest baseline, recurrent networks, and fully connected networks, reporting higher accuracy metrics for every instrument. Attention also produces interpretable outputs by highlighting the specific time segments the model uses for each label. A sympathetic reader would care because real music often contains overlapping instruments and weak labels are far cheaper to obtain than frame-by-frame annotations. If the claim holds, attention offers a practical route to scaling instrument recognition without requiring expensive strong supervision.

Core claim

The paper claims that an attention mechanism applied to audio features for multi-label instrument recognition produces an overall improvement in classification accuracy across all 20 instruments on the OpenMIC dataset. The same mechanism lets the model focus on time segments relevant to each instrument label even when trained solely on weak clip-level presence/absence annotations, yielding both higher performance and more interpretable results than baseline random forests, recurrent networks, or fully connected networks.

What carries the argument

The attention mechanism that computes per-instrument weights over time segments in the audio feature sequence.

If this is right

  • Attention models outperform the random forest, RNN, and fully connected baselines on every instrument in OpenMIC.
  • Attention produces interpretable outputs by indicating which time segments support each instrument label.
  • The approach works with weak labels alone and does not require per-frame annotations.
  • Performance gains appear consistently across the full set of 20 instruments.

Where Pith is reading between the lines

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

  • The same attention layer could be added to models for other audio tasks that rely on weak labels, such as sound event detection.
  • If the attended segments prove reliable, they might serve as a cheap source of pseudo frame-level labels for further training.
  • Testing the attention model on smaller strongly labeled sets like MedleyDB would show whether the benefit holds when frame information is already available.

Load-bearing premise

The attention mechanism can reliably locate instrument-relevant time segments using only weak clip-level labels without per-frame annotations or extra supervision.

What would settle it

Retraining the attention model on OpenMIC and finding no accuracy gain over the recurrent baseline, or finding that the attended segments show no better alignment with actual instrument activity than random segments on a dataset with frame-level labels, would falsify the claim.

read the original abstract

While the automatic recognition of musical instruments has seen significant progress, the task is still considered hard for music featuring multiple instruments as opposed to single instrument recordings. Datasets for polyphonic instrument recognition can be categorized into roughly two categories. Some, such as MedleyDB, have strong per-frame instrument activity annotations but are usually small in size. Other, larger datasets such as OpenMIC only have weak labels, i.e., instrument presence or absence is annotated only for long snippets of a song. We explore an attention mechanism for handling weakly labeled data for multi-label instrument recognition. Attention has been found to perform well for other tasks with weakly labeled data. We compare the proposed attention model to multiple models which include a baseline binary relevance random forest, recurrent neural network, and fully connected neural networks. Our results show that incorporating attention leads to an overall improvement in classification accuracy metrics across all 20 instruments in the OpenMIC dataset. We find that attention enables models to focus on (or `attend to') specific time segments in the audio relevant to each instrument label leading to interpretable results.

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

3 major / 2 minor

Summary. The manuscript proposes an attention mechanism to improve multi-label musical instrument recognition on weakly labeled data such as the OpenMIC dataset (20 instruments). It compares the attention model against baselines including a binary relevance random forest, RNN, and FCNN, claiming that attention yields an overall improvement in classification accuracy metrics across all instruments while enabling interpretable focus on relevant time segments from clip-level labels alone.

Significance. If the reported gains can be shown to arise specifically from the attention component's ability to identify instrument-relevant segments under weak supervision (rather than from added capacity), the approach could usefully extend attention-based methods for polyphonic audio tasks with limited annotations. The interpretability claim is a secondary potential contribution for MIR applications.

major comments (3)
  1. [Abstract] Abstract: the central empirical claim states an 'overall improvement in classification accuracy metrics across all 20 instruments' yet supplies no numerical values, per-instrument scores, aggregate metrics, error bars, or statistical tests, so the magnitude and reliability of the improvement cannot be assessed.
  2. [Experiments] Experiments/Results (comparison to baselines): no ablation is presented that holds base architecture and parameter count fixed while toggling only the attention module, leaving open the possibility that gains arise from increased model capacity rather than from attention learning relevant segments from weak labels (directly addressing the stress-test concern).
  3. [Method] Method section: the integration of attention with clip-level weak labels is described at a high level only; without details on the attention formulation, loss, or how per-frame relevance is learned without frame-level supervision, it is not possible to verify that the mechanism operates as claimed under the weakest assumption.
minor comments (2)
  1. [Abstract] Abstract: 'classification accuracy metrics' is used without specifying the exact measures (e.g., micro/macro F1, AUC, precision@K) or how multi-label predictions are thresholded.
  2. The manuscript would benefit from explicit statements of training procedure, data splits, hyperparameter selection, and whether any post-hoc choices were made on the test set.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major point below and will incorporate revisions to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central empirical claim states an 'overall improvement in classification accuracy metrics across all 20 instruments' yet supplies no numerical values, per-instrument scores, aggregate metrics, error bars, or statistical tests, so the magnitude and reliability of the improvement cannot be assessed.

    Authors: We agree that the abstract should include concrete numerical support for the improvement claim. In the revised manuscript we will add the key aggregate metrics (e.g., mean F1-score across instruments) and note the range of per-instrument gains, while directing readers to the results section for full tables, error bars, and any statistical comparisons. revision: yes

  2. Referee: [Experiments] Experiments/Results (comparison to baselines): no ablation is presented that holds base architecture and parameter count fixed while toggling only the attention module, leaving open the possibility that gains arise from increased model capacity rather than from attention learning relevant segments from weak labels (directly addressing the stress-test concern).

    Authors: The referee correctly identifies the absence of a controlled ablation. We will add an ablation experiment that compares the attention model against an otherwise identical architecture (same backbone, same parameter budget) with the attention module removed or replaced by a simple pooling layer, thereby isolating the contribution of the attention component under weak supervision. revision: yes

  3. Referee: [Method] Method section: the integration of attention with clip-level weak labels is described at a high level only; without details on the attention formulation, loss, or how per-frame relevance is learned without frame-level supervision, it is not possible to verify that the mechanism operates as claimed under the weakest assumption.

    Authors: We acknowledge that the current method description is high-level. The revised manuscript will expand this section with the explicit attention equations, the precise loss formulation (including how clip-level labels supervise frame-level attention weights), and a step-by-step account of how relevance is learned without frame annotations. revision: yes

Circularity Check

0 steps flagged

No circularity in empirical model comparison

full rationale

The paper presents an empirical study comparing an attention-augmented neural network against baselines (random forest, RNN, FCNN) on the OpenMIC dataset for multi-label instrument recognition. No mathematical derivations, equations, or first-principles results are described that could reduce to self-definition, fitted inputs renamed as predictions, or self-citation chains. The reported accuracy gains are experimental outcomes whose independence from the listed circularity patterns cannot be challenged from the provided text; the work is self-contained as a standard ML ablation study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no model equations, training details, or parameter counts are supplied, so the ledger cannot enumerate concrete free parameters, axioms, or invented entities. The central claim implicitly assumes that standard neural-network training on weak labels plus attention is sufficient to produce the reported gains.

pith-pipeline@v0.9.0 · 5719 in / 1104 out tokens · 14110 ms · 2026-05-24T23:59:48.422594+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

55 extracted references · 55 canonical work pages · 3 internal anchors

  1. [1]

    An Attention Mechanism for Musical Instrument Recognition

    INTRODUCTION Musical instruments, both acoustic and electronic, are nec- essary tools to create music. Most musical pieces comprise of a combination of multiple musical instruments resulting in a mixture with unique timbre characteristics. Humans are fairly adept at recognizing musical instruments in the music they hear. Recognizing instruments automatica...

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  2. [2]

    We refer to [15,26] for a review of literature in single instrument and monophonic instrument recognition

    RELATED WORK 2.1 Musical Instrument Recognition Instrument recognition in audio containing a single instru- ment can refer to both recognition from isolated notes or recognition from solo recordings of pieces. We refer to [15,26] for a review of literature in single instrument and monophonic instrument recognition. Current research has focused on instrume...

    work page

  3. [3]

    2.1, we introduced research on instrument recogni- tion in polyphonic, multi-timbral music

    DATA CHALLENGE In Sec. 2.1, we introduced research on instrument recogni- tion in polyphonic, multi-timbral music. One theme that emerges is that with almost every new publication, a new dataset is released by the authors in an effort to address issues with previous ones. While releasing new datasets is highly encouraged and vital for research in MIR in g...

    work page

  4. [4]

    4.1 Pre-Processing As mentioned in Sect

    METHOD Before describing the model details, we provide a formaliza- tion of our approach to the instrument recognition problem in weakly labeled data. 4.1 Pre-Processing As mentioned in Sect. 3, the OpenMIC dataset consists of 10 s audio clips, each labeled with the presence or absence of one or more of 20 instrument labels. For each audio file in the data...

    work page

  5. [5]

    5.1 Dataset We use the OpenMIC dataset for the experiments in this paper

    EV ALUATION In this section we describe the experimental setup including the dataset, the baseline methods, and evaluation metrics. 5.1 Dataset We use the OpenMIC dataset for the experiments in this paper. In addition to the audio and label annotations, the data repository contains pre-computed features extracted from the publicly available VGG-ish model ...

    work page

  6. [6]

    A binary-relevance transformation is applied to convert the multi-label classification task into 20 independent binary classification tasks [39]

    RF_BR: This model is the baseline random forest model in [17]. A binary-relevance transformation is applied to convert the multi-label classification task into 20 independent binary classification tasks [39]

    work page

  7. [7]

    Here, the input features of dimension 10× 128 are flattened into a single feature vector for classification

    FC: A 3-layer fully connected network trained to predict the presence or absence of all instruments for a given data instance. Here, the input features of dimension 10× 128 are flattened into a single feature vector for classification. Dropout is used for regularization and the Leaky ReLU ( 0.01 slope) is used. The model has 986772 parameters

    work page

  8. [8]

    FC_T uses the same embedding layer as ATT

    FC_T: This model serves as an ablation study to ob- serve the benefits of the attention mechanism. FC_T uses the same embedding layer as ATT. However, the aggregation of predictions in time is simply per- formed with average-pooling. The model has 52116 parameters

    work page

  9. [9]

    The model processes the input features and produces a single embedding which is then fed to a classifier for all 20 instruments

    RNN: A 3-layer bi-directional gated recurrent unit model with 64 hidden units per direction. The model processes the input features and produces a single embedding which is then fed to a classifier for all 20 instruments. The model has 226068 parameters. Source code for the Pytorch implementation of the neu- ral network models is publicly available. 1 For ...

    work page

  10. [10]

    Additionally, we compare the instrument-wise F1-score for each model in Figure 3

    RESULTS AND DISCUSSION Figure 2 shows the overall performance of ATT compared to the baseline models with box plots for the macro- averaged precision, recall, and F1-score. Additionally, we compare the instrument-wise F1-score for each model in Figure 3. Note that we only show the mean instrument-wise F1-score across 10 seeds in Figure 3 for improved visi...

    work page

  11. [11]

    This calls for a paradigm shift in the ap- proaches towards supervised learning approaches better suited for weakly labeled data

    CONCLUSION Weakly labeled datasets for instrument recognition in poly- phonic music are easier to develop or annotate than strongly labeled datasets. This calls for a paradigm shift in the ap- proaches towards supervised learning approaches better suited for weakly labeled data. We formulate the instru- ment recognition task as a MIML problem and introduc...

    work page

  12. [12]

    We thank them for their generous support and meaningful dis- cussions

    ACKNOWLEDGEMENTS This research is partially funded by Gracenote, Inc. We thank them for their generous support and meaningful dis- cussions. We also thank Nvidia Corporation for their dona- tion of a Titan V awarded as part of the GPU grant program

    work page

  13. [13]

    Sound event detection using spatial features and convo- lutional recurrent neural network

    Sharath Adavanne, Pasi Pertilä, and Tuomas Virtanen. Sound event detection using spatial features and convo- lutional recurrent neural network. In Proc. of the IEEE International Conference on Acoustics, Speech and Sig- nal Processing (ICASSP), pages 771–775, New Orleans, LA, USA, 2017

    work page 2017

  14. [14]

    Neural machine translation by jointly learning to align and translate

    Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Ben- gio. Neural machine translation by jointly learning to align and translate. In Proc. of the International Confer- ence on Learning Representations, (ICLR), San Diego, CA, USA, 2015

    work page 2015

  15. [15]

    Multilabel classification with label correlations and missing labels

    Wei Bi and James Kwok. Multilabel classification with label correlations and missing labels. In Proc. of the AAAI Conference on Artificial Intelligence, pages 1680– 1686, Québec City, Québec, Canada, 2014

    work page 2014

  16. [16]

    Medleydb: A multitrack dataset for annotation- intensive MIR research

    Rachel M Bittner, Justin Salamon, Mike Tierney, Matthias Mauch, Chris Cannam, and Juan Pablo Bello. Medleydb: A multitrack dataset for annotation- intensive MIR research. In Proc. of the International Society for Music Information Retrieval Conference (IS- MIR), pages 155–160, Taipei, Taiwan, 2014

    work page 2014

  17. [17]

    A comparison of sound segregation techniques for predominant instrument recognition in musical audio signals

    Juan J Bosch, Jordi Janer, Ferdinand Fuhrmann, and Perfecto Herrera. A comparison of sound segregation techniques for predominant instrument recognition in musical audio signals. In Proc. of the International So- ciety for Music Information Retrieval Conference (IS- MIR), pages 559–564, Porto, Portugal, 2012

    work page 2012

  18. [18]

    Convolutional recurrent neural networks for polyphonic sound event detection

    Emre Cakır, Giambattista Parascandolo, Toni Heittola, Heikki Huttunen, and Tuomas Virtanen. Convolutional recurrent neural networks for polyphonic sound event detection. IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP), 25(6):1291–1303, 2017

    work page 2017

  19. [19]

    Learning a deep convnet for multi-label classification with partial labels

    Thibaut Durand, Nazanin Mehrasa, and Greg Mori. Learning a deep convnet for multi-label classification with partial labels. In Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 647–657, Long Beach, CA, USA, 2019

    work page 2019

  20. [20]

    A review of multi- instance learning assumptions

    James Foulds and Eibe Frank. A review of multi- instance learning assumptions. The Knowledge Engi- neering Review, 25(1):1–25, 2010

    work page 2010

  21. [21]

    Automatic musical instrument recognition from polyphonic music audio signals

    Ferdinand Fuhrmann. Automatic musical instrument recognition from polyphonic music audio signals. PhD thesis, Universitat Pompeu Fabra, 2012

    work page 2012

  22. [22]

    Gemmeke, Daniel P

    Jort F. Gemmeke, Daniel P. W. Ellis, Dylan Freedman, Aren Jansen, Wade Lawrence, R. Channing Moore, Manoj Plakal, and Marvin Ritter. Audio set: An on- tology and human-labeled dataset for audio events. In Proc. of the IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP) , pages 776–780, New Orleans, LA, USA, 2017

    work page 2017

  23. [23]

    Mixing se- crets: A multi-track dataset for instrument detection in polyphonic music

    Siddharth Gururani and Alexander Lerch. Mixing se- crets: A multi-track dataset for instrument detection in polyphonic music. In Late Breaking Demo (Extended Abstract), Proc. of the International Society for Mu- sic Information Retrieval Conference (ISMIR), Suzhou, China, 2017

    work page 2017

  24. [24]

    Instrument activity detection in polyphonic music using deep neural networks

    Siddharth Gururani, Cameron Summers, and Alexan- der Lerch. Instrument activity detection in polyphonic music using deep neural networks. In Proc. of the Inter- national Society for Music Information Retrieval Con- ference (ISMIR), pages 569–576, Paris, France, 2018

    work page 2018

  25. [25]

    Deep convolutional neural networks for predominant instru- ment recognition in polyphonic music

    Yoonchang Han, Jaehun Kim, and Kyogu Lee. Deep convolutional neural networks for predominant instru- ment recognition in polyphonic music. IEEE/ACM Transactions on Audio, Speech and Language Process- ing (TASLP), 25(1):208–221, 2017

    work page 2017

  26. [26]

    Sparse feature learning for instrument identi- fication: Effects of sampling and pooling methods

    Yoonchang Han, Subin Lee, Juhan Nam, and Kyogu Lee. Sparse feature learning for instrument identi- fication: Effects of sampling and pooling methods. The Journal of the Acoustical Society of America , 139(5):2290–2298, 2016

    work page 2016

  27. [27]

    Automatic classification of musical instrument sounds

    Perfecto Herrera-Boyer, Geoffroy Peeters, and Shlomo Dubnov. Automatic classification of musical instrument sounds. Journal of New Music Research, 32(1):3–21, 2003

    work page 2003

  28. [28]

    Shawn Hershey, Sourish Chaudhuri, Daniel P. W. El- lis, Jort F. Gemmeke, Aren Jansen, Channing Moore, Manoj Plakal, Devin Platt, Rif A. Saurous, Bryan Sey- bold, Malcolm Slaney, Ron Weiss, and Kevin Wilson. CNN architectures for large-scale audio classification. In Proc. of the IEEE International Conference on Acous- tics, Speech and Signal Processing (ICA...

    work page 2017

  29. [29]

    Openmic-2018: An open dataset for multiple instru- ment recognition

    Eric Humphrey, Simon Durand, and Brian McFee. Openmic-2018: An open dataset for multiple instru- ment recognition. In Proc. of the International Society for Music Information Retrieval Conference (ISMIR), pages 438–444, Paris, France, 2018

    work page 2018

  30. [30]

    Mul- titask learning for frame-level instrument recognition

    Yun-Ning Hung, Yi-An Chen, and Yi-Hsuan Yang. Mul- titask learning for frame-level instrument recognition. In Proc. of the IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP) , pages 381–385, Brighton, UK, 2019

    work page 2019

  31. [31]

    Frame-level instru- ment recognition by timbre and pitch

    Yun-Ning Hung and Yi-Hsuan Yang. Frame-level instru- ment recognition by timbre and pitch. In Proc. of the International Society for Music Information Retrieval Conference (ISMIR) , pages 135–142, Paris, France, 2018

    work page 2018

  32. [32]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Proc. of the International Conference on Learning Representations, (ICLR), San Diego, CA, USA, 2015

    work page 2015

  33. [33]

    Instrument iden- tification in polyphonic music: Feature weighting to minimize influence of sound overlaps

    Tetsuro Kitahara, Masataka Goto, Kazunori Komatani, Tetsuya Ogata, and Hiroshi G Okuno. Instrument iden- tification in polyphonic music: Feature weighting to minimize influence of sound overlaps. EURASIP Jour- nal on Applied Signal Processing , 2007(1):155–155, 2007

    work page 2007

  34. [34]

    Audio set classification with attention model: A probabilistic perspective

    Qiuqiang Kong, Yong Xu, Wenwu Wang, and Mark Plumbley. Audio set classification with attention model: A probabilistic perspective. In Proc. of the IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 316–320, Calgary, Canada, 2018

    work page 2018

  35. [35]

    Plumbley

    Qiuqiang Kong, Changsong Yu, Turab Iqbal, Yong Xu, Wenwu Wang, and Mark D. Plumbley. Weakly labelled audioset classification with attention neural networks. CoRR, abs/1903.00765, 2019

    work page arXiv 1903

  36. [36]

    Audio event detection using weakly labeled data

    Anurag Kumar and Bhiksha Raj. Audio event detection using weakly labeled data. In Proc. of the 24th ACM International Conference on Multimedia (ACMMM) , pages 1038–1047, Amsterdam, The Netherlands, 2016

    work page 2016

  37. [37]

    Automatic Instrument Recognition in Polyphonic Music Using Convolutional Neural Networks

    Peter Li, Jiyuan Qian, and Tian Wang. Automatic in- strument recognition in polyphonic music using convo- lutional neural networks. CoRR, abs/1511.05520, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  38. [38]

    Extended Playing Techniques: The Next Mile- stone in Musical Instrument Recognition

    Vincent Lostanlen, Joakim Andén, and Mathieu La- grange. Extended Playing Techniques: The Next Mile- stone in Musical Instrument Recognition. In Proc. of the International Conference on Digital Libraries for Musicology (DLfM), pages 1–10, Paris, France, 2018

    work page 2018

  39. [39]

    Adaptive pooling operators for weakly labeled sound event detection

    Brian McFee, Justin Salamon, and Juan Pablo Bello. Adaptive pooling operators for weakly labeled sound event detection. IEEE/ACM Transactions on Au- dio, Speech, and Language Processing (TASLP) , 26(11):2180–2193, 2018

    work page 2018

  40. [40]

    Correlative multi-label video annotation

    Guo-Jun Qi, Xian-Sheng Hua, Yong Rui, Jinhui Tang, Tao Mei, and Hong-Jiang Zhang. Correlative multi-label video annotation. In Proc. of the ACM International Conference on Multimedia (ACMMM) , pages 17–26, Augsburg, Germany, 2007

    work page 2007

  41. [41]

    Colin Raffel and Daniel P. W. Ellis. Feed-forward net- works with attention can solve some long-term memory problems. CoRR, abs/1512.08756, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  42. [42]

    Very deep convolutional networks for large-scale image recogni- tion

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recogni- tion. In Proc. of the International Conference on Learn- ing Representations, (ICLR) , San Diego, CA, USA, 2015

    work page 2015

  43. [43]

    Plumbley

    Dan Stowell, Dimitrios Giannoulis, Emmanouil Bene- tos, Mathieu Lagrange, and Mark D. Plumbley. Detec- tion and classification of acoustic scenes and events. IEEE Transactions on Multimedia, 17(10):1733–1746, 2015

    work page 2015

  44. [44]

    Instrudive: A Music Visualization System Based on Automatically Recognized Instrumentation

    Takumi Takahashi, Satoru Fukayama, and Masataka Goto. Instrudive: A Music Visualization System Based on Automatically Recognized Instrumentation. In Proc. of the International Society for Music Information Retrieval Conference (ISMIR), pages 561–568, Paris, France, 2018

    work page 2018

  45. [45]

    Learning features of music from scratch

    John Thickstun, Zaïd Harchaoui, and Sham Kakade. Learning features of music from scratch. In Proc. of the International Conference on Learning Representations, (ICLR), Toulon, France, 2017

    work page 2017

  46. [46]

    Multi-label classifica- tion of music into emotions

    Konstantinos Trohidis, Grigorios Tsoumakas, George Kalliris, and Ioannis P Vlahavas. Multi-label classifica- tion of music into emotions. InProc. of the International Society for Music Information Retrieval Conference (IS- MIR), pages 325–330, Philadelphia, PA, USA, 2008

    work page 2008

  47. [47]

    Attention is All you Need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is All you Need. In Advances in Neural Information Processing Systems (NeurIPS), pages 5998–6008. Curran Associates, Inc., Long Beach, CA, USA, 2017

    work page 2017

  48. [48]

    From labeled to unlabeled data – on the data challenge in automatic drum transcription

    Chih-Wei Wu and Alexander Lerch. From labeled to unlabeled data – on the data challenge in automatic drum transcription. In Proc. of the International Society for Music Information Retrieval Conference (ISMIR), Paris, France, 2018

    work page 2018

  49. [49]

    Deep sets

    Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov, and Alexan- der J Smola. Deep sets. In Advances in Neural Informa- tion Processing Systems (NeurIPS), pages 3391–3401. Curran Associates, Inc., Long Beach, CA, USA, 2017

    work page 2017

  50. [50]

    Joint multi-label multi-instance learning for image classification

    Zheng-Jun Zha, Xian-Sheng Hua, Tao Mei, Jingdong Wang, Guo-Jun Qi, and Zengfu Wang. Joint multi-label multi-instance learning for image classification. InProc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1–8, Anchorage, AK, USA, 2008

    work page 2008

  51. [51]

    Binary relevance for multi-label learning: an overview

    Min-Ling Zhang, Yu-Kun Li, Xu-Ying Liu, and Xin Geng. Binary relevance for multi-label learning: an overview. Frontiers of Computer Science, 12(2):191– 202, 2018

    work page 2018

  52. [52]

    Multi-label learning by exploiting label dependency

    Min-Ling Zhang and Kun Zhang. Multi-label learning by exploiting label dependency. In Proc. of the ACM International Conference on Knowledge Discovery and Data Mining (ACM SIGKDD), pages 999–1008, Wash- ington, DC, USA, 2010

    work page 2010

  53. [53]

    Neural networks for multi-instance learning

    Zhi-Hua Zhou and Min-Ling Zhang. Neural networks for multi-instance learning. In Proc. of the Interna- tional Conference on Intelligent Information Technol- ogy, pages 455–459, Beijing, China, 2002

    work page 2002

  54. [54]

    Multi-instance multi-label learning with application to scene classifica- tion

    Zhi-Hua Zhou and Min-Ling Zhang. Multi-instance multi-label learning with application to scene classifica- tion. In Advances in Neural Information Processing Sys- tems (NeurIPS), pages 1609–1616. Curran Associates, Inc., Vancouver, BC, Canada, 2007

    work page 2007

  55. [55]

    Multi-instance multi-label learning

    Zhi-Hua Zhou, Min-Ling Zhang, Sheng-Jun Huang, and Yu-Feng Li. Multi-instance multi-label learning. Artificial Intelligence, 176(1):2291 – 2320, 2012

    work page 2012