Study of the Properties of LLM Embeddings: Continious Interpretation and Robustness of Dense Point Clouds

Student: Zubrey Aleksandr

Educational Programme: Math of Machine Learning (Master)

Year of Graduation: 2024

In this study we research the resilience of large language models to the noise added to input token embeddings. Added noise can also be interpret as tokens being sampled from a continuous probability distribution rather than discrete point cloud. Our experiments confirm that existing pretrained models tolerate small amount of noise and can quickly adapt to larger noise by finetuning without changing the original positions of embeddings. We also found out that in some cases a small noise even allows to increase the output quality. Our findings give a new look on structure of LLMs and their embeddings.

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses