Single-Branch Architectures for Recommendation Marta Moscati 0 Johannes Kepler University Linz , Linz , Austria 2026

Single-branch architectures have been proven efective for several multimodal learning tasks. In this talk, after reviewing the use of single-branch architectures in multimodal learning, I describe their use in multimodal recommendation, showing how they allow to address missing-modality and cold-start scenarios. I then describe the use of single-branch architectures in collaborative filtering, showing how they allow to reduce the number of model parameters without substantially afecting the quality of recommendations.

 eol>Multimodal Learning Multimodal Recommender Systems Collaborative Filtering -

batch userinterac ons batch batch contras ve loss recommloesnsda on (a) SiBraR [ 21 ]: Single-branch multimodal RS. (b) CoBraR [26]: Single-branch for CF. provides more accurate recommendations compared to well-established CF and multimodal RS [ 21 ], and that the performance improvement is to be attributed to the simultaneous use of the single-branch architecture and the contrastive loss [25].

Single-Branch Architectures in Collaborative Filtering. Modern CF algorithms [27] rely on representing users and items in a joint embedding space where these are close if corresponding to a positive interaction. Motivated by the analogy between this approach and the objective of reducing the modality gap in multimodal learning, as well as by the observation that weight sharing allows to simultaneously reduce the modality gap and to reduce the number of model parameters, we proposed the use of a single-branch architecture for CF. The resulting model, named Collaborative Branch Recommender (CoBraR [26]), is depicted in Figure 1b. The architecture and training paradigm of CoBraR are similar to the well-established two-branch CF algorithm DMF [28], where the user and item representations are obtained by converting their sparse interaction vectors, i.e., the corresponding row or column in the user–item interaction matrix, to dense vectors by mean of two multi-layer perceptrons (MLP) - one MLP for the user and one MLP for the item. The novelty of CoBraR is that the same MLP is used to encode both users and items. We showed that CoBraR reduces the number of model parameters compared to its two-tower counterpart DMF, improving beyond-accuracy aspects of recommendation quality without compromising accuracy.

Future Applications of Single-Branch Architectures to Recommendation. Both SiBraR and CoBraR translate single-branch architectures to RS and open the way for future applications. First, although SiBraR has been proven efective for tackling scenarios where one modality is missing at inference time, its architecture has not yet been evaluated on scenarios where one modality is also missing during training. As for CoBraR, its architecture and training paradigm are based on DMF. Therefore, there is no evidence yet of whether using weight sharing between users and items would also be efective with other loss functions for recommendation, or for other two-branch CF models as well, such as NeuMF [29]. Finally, SiBraR and CoBraR could be combined in a model where a same multimodal single-branch is applied to all multimodal data of both users and items, including interactions and side-information.

 1. Acknowledgments

I am grateful to Markus Schedl, Shah Nawaz, Christian Ganhör, and Anna Hausberger for the fruitful collaboration and invaluable discussions that led to the development SiBraR and CoBraR. This research was funded in whole or in part by the Austrian Science Fund (FWF) https://doi.org/10.55776/P33526, https://doi.org/10.55776/DFH23, https://doi.org/10.55776/COE12, https://doi.org/10.55776/P36413. No generative AI tool was used during the preparation of this work.

 Declaration on Generative AI

The author have not employed any Generative AI tools. [24] A. van den Oord, Y. Li, O. Vinyals, Representation learning with contrastive predictive coding, 2019. arXiv:1807.03748. [25] C. Ganhör, M. Moscati, A. Hausberger, S. Nawaz, M. Schedl, Single-branch network architectures to close the modality gap in multimodal recommendation, ACM Transactions on Recommender Systems (2025). [26] M. Moscati, S. Nawaz, M. Schedl, Parameter-eficient single collaborative branch for recommendation, in: Proc. of ACM RecSys, 2025. [27] S. Zhang, Y. Tay, L. Yao, A. Sun, C. Zhang, Deep learning for recommender systems, in: F. Ricci,

L. Rokach, B. Shapira (Eds.), Recommender Systems Handbook, Springer US, New York, NY, 2022. [28] H.-J. Xue, X.-Y. Dai, J. Zhang, S. Huang, J. Chen, Deep matrix factorization models for recommender systems, in: Proc. of IJCAI, 2017. [29] X. He, L. Liao, H. Zhang, L. Nie, X. Hu, T.-S. Chua, Neural collaborative filtering, in: Proc. of WWW, 2017.

 [1] T. Baltrušaitis , C. Ahuja , L.-P. Morency, Multimodal machine learning: A survey and taxonomy , IEEE Transactions on Pattern Analysis and Machine Intelligence 41 ( 2019 ). [2] P. Xu , X. Zhu , D. A. Clifton , Multimodal learning with transformers: A survey , IEEE Transactions on Pattern Analysis and Machine Intelligence ( 2023 ). [3] Q. Liu , J. Hu , Y. Xiao , X. Zhao , J. Gao , W. Wang , Q. Li , J. Tang , Multimodal recommender systems: A survey , ACM Computing Surveys 57 ( 2024 ). [4] A. Radford , J. W. Kim , C. Hallacy , A. Ramesh , G. Goh, S. Agarwal , G. Sastry , A. Askell , P. Mishkin , J. Clark , G. Krueger , I. Sutskever , Learning transferable visual models from natural language supervision , in: Proc. of ICML , 2021 . [5] S. Reed , Z. Akata , H. Lee , B. Schiele , Learning deep representations of fine-grained visual descriptions , in: Proc. of IEEE CVPR , 2016 . [6] A. Nagrani , S. Albanie , A. Zisserman , Seeing voices and hearing faces: Cross-modal biometric matching , in: Proc. of IEEE CVPR , 2018 . [7] A. Nagrani , S. Albanie , A. Zisserman , Learnable pins: Cross-modal embeddings for person identity , in: Proc. of ECCV , 2018 . [8] M. S. Saeed , M. H. Khan , S. Nawaz , M. H. Yousaf , A. Del Bue , Fusion and orthogonal projection for improved face-voice association , in: Proc. of ICASSP , 2022 . [9] O. Arshad , I. Gallo, S. Nawaz , A. Calefati , Aiding intra-text representations with visual context for multimodal named entity recognition , in: Proc. of IEEE ICDAR , 2019 . [10] A. Vaswani , N. Shazeer , N. Parmar , J. Uszkoreit , L. Jones , A. N. Gomez , L. u. Kaiser, I. Polosukhin , Attention is all you need , in: Proc. of NeurIPS , 2017 . [11] J. Lu , D. Batra , D. Parikh , S. Lee , Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks , in: Proc. of NeurIPS , 2019 . [12] H. Tan , M. Bansal , Lxmert: Learning cross-modality encoder representations from transformers , in: Proc. of EMNLP , 2019 . [13] A. Hannan , M. A. Manzoor , S. Nawaz , M. I. Liaqat , M. Schedl , M. Noman, PAEFF: Precise Alignment and Enhanced Gated Feature Fusion for Face-Voice Association , in : Proc. of Interspeech , 2025 . [14] M. S. Saeed , S. Nawaz , M. H. Khan , M. Z. Zaheer , K. Nandakumar , M. H. Yousaf , A. Mahmood , Single-branch network for multimodal training , in: Proc. of ICASSP , 2023 . [15] M. Tschannen , B. Mustafa , N. Houlsby , Clippo: Image-and-language understanding from pixels only , in: Proc. of IEEE/CVF CVPR , 2023 . [16] M. I. Liaqat , S. Nawaz , M. Z. Zaheer , M. S. Saeed , H. Sajjad , T. D. Schepper , K. Nandakumar , M. H. Khan , I. Gallo, M. Schedl , Chameleon: A multimodal learning framework robust to missing modalities , International Journal of Multimedia Information Retrieval 14 ( 2025 ). [17] J. Geiger , M. Moscati , S. Nawaz , M. Schedl, Music4all a+a: A multimodal dataset for music information retrieval tasks , in: Proc. of CBMI , 2025 . [18] M. Ma , J. Ren , L. Zhao , D. Testuggine , X. Peng , Are multimodal transformers robust to missing modality? , in: Proc. of IEEE/CVF CVPR , 2022 . [19] D. Malitesta , G. Cornacchia, C. Pomo , F. A. Merra , T. Di Noia , E. Di Sciascio , Formalizing multimedia recommendation through multimodal deep learning , ACM Transactions on Recommender Systems 3 ( 2025 ). [20] F. Ricci , L. Rokach , B. Shapira (Eds.), Recommender Systems Handbook , Springer US, New York, NY, 2022 . [21] C. Ganhör * , M. Moscati* , A. Hausberger , S. Nawaz , M. Schedl , A multimodal single-branch embedding network for recommendation in cold-start and missing modality scenarios , in: Proc. of ACM RecSys, *Equal Contributions , 2024 . [22] S. Rendle , C. Freudenthaler , Z. Gantner , L. Schmidt-Thieme , Bpr: Bayesian personalized ranking from implicit feedback , in: Proc. of UAI , 2009 . [23] Z. Yuhui , Y. Wada , H. Waida , K. Goto , Y. Hino , T. Kanamori, Deep clustering with a constraint for topological invariance based on symmetric infonce , Neural Computation 35 ( 2023 ).