Aspects of Similarity in Natural and Artificial Minds
Humans construct useful representations to support adaptive behavior in high-dimensional sensory environments. Understanding the structure of those representations has been at the center of decades of psychological research. Human similarity judgments provide an effective way for characterizing the structure of representations by densely mapping the perceived relations between pairs of stimuli. This idea propelled a fruitful research program that yielded many classic results in the literature such as Shepard’s universal law of generalization, Tversky’s feature model, the helical representation of pitch, and multidimensional scaling analysis.
Despite its success, the similarity research program is not complete, in part because central results such as the universal law of generalization are based on low-dimensional artificial stimuli and small sample sizes that severely limit their generalizability, and in part because recent methodological advances such as the development of large language models, contrastive training, and adaptive experiment designs unlock new sets of questions that were simply not feasible before.
In this set of projects I combine the similarity-based approach with large-scale adaptive experiments, modern machine learning, and Bayesian modeling to address questions pertaining to three key aspects of representations: (i) Scaling: To what extent does representational structure generalize to naturalistic and ecologically valid tasks? (Marjieh et al., 2024; Marjieh et al., 2024) (ii) Alignment: How much of our cognitive representations can be recovered from language? (Marjieh et al., 2024; Marjieh et al., 2023; Marjieh et al., 2025) and (iii) Abstraction: What form of similarity can support learning abstract representations in humans and machines? (Marjieh et al., 2025)
References
2025
What is a Number, That a Large Language Model May Know It?
Raja Marjieh, Veniamin Veselovsky, Thomas L. Griffiths, and 1 more author
@article{marjieh2025numberlargelanguagemodel,title={What is a Number, That a Large Language Model May Know It?},author={Marjieh, Raja and Veselovsky, Veniamin and Griffiths, Thomas L. and Sucholutsky, Ilia},year={2025},journal={arXiv preprint arXiv:2502.01540},}
Learning Human-Aligned Representations with Contrastive Learning and Generative Similarity
Raja Marjieh, Sreejan Kumar, Declan Campbell, and 4 more authors
@article{marjieh2025learninghumanalignedrepresentationscontrastive,title={Learning Human-Aligned Representations with Contrastive Learning and Generative Similarity},author={Marjieh, Raja and Kumar, Sreejan and Campbell, Declan and Zhang, Liyi and Bencomo, Gianluca and Snell, Jake and Griffiths, Thomas L.},year={2025},journal={arXiv preprint arXiv:2405.19420},url={https://arxiv.org/abs/2405.19420},}
2024
The universal law of generalization holds for naturalistic stimuli.
Raja Marjieh, Nori Jacoby, Joshua C Peterson, and 1 more author
Journal of Experimental Psychology: General. Selected as Editor’s Choice , 2024
@article{marjieh2024universal,title={The universal law of generalization holds for naturalistic stimuli.},author={Marjieh, Raja and Jacoby, Nori and Peterson, Joshua C and Griffiths, Thomas L},journal={Journal of Experimental Psychology: General},volume={153},number={3},pages={573},year={2024},publisher={American Psychological Association},url={https://psycnet.apa.org/record/2024-56632-001},}
Pitch is Not a Helix: Probing the Structure of Musical Pitch Across Tasks and Experience
Raja Marjieh, Thomas L. Griffiths, and Nori Jacoby
@article{marjieh2023pitch,author={Marjieh, Raja and Griffiths, Thomas L. and Jacoby, Nori},title={Pitch is Not a Helix: Probing the Structure of Musical Pitch Across Tasks and Experience},year={2024},publisher={Cold Spring Harbor Laboratory},journal={bioRxiv},}
Large language models predict human sensory judgments across six modalities
Raja Marjieh, Ilia Sucholutsky, Pol Rijn, and 2 more authors
@article{marjieh2024large,title={Large language models predict human sensory judgments across six modalities},author={Marjieh, Raja and Sucholutsky, Ilia and van Rijn, Pol and Jacoby, Nori and Griffiths, Thomas L},journal={Scientific Reports},volume={14},number={1},pages={21445},year={2024},publisher={Nature Publishing Group UK London},url={https://www.nature.com/articles/s41598-024-72071-1},}
2023
Words are all you need? Language as an approximation for human similarity judgments
Raja Marjieh, Pol Van Rijn, Ilia Sucholutsky, and 4 more authors
In The Eleventh International Conference on Learning Representations, 2023
@inproceedings{marjieh2023words,title={Words are all you need? Language as an approximation for human similarity judgments},author={Marjieh, Raja and Rijn, Pol Van and Sucholutsky, Ilia and Sumers, Theodore and Lee, Harin and Griffiths, Thomas L. and Jacoby, Nori},booktitle={The Eleventh International Conference on Learning Representations},year={2023},}
