Santiago Acevedo

Santiago Acevedo, Alessandro Laio, Marco Baroni

We study how syntactic and semantic information is encoded in inner layer representations of Large Language Models (LLMs), focusing on the very large DeepSeek-V3. We find that, by averaging hidden-representation vectors of sentences sharing syntactic structure or meaning, we obtain vectors that capture a significant proportion of the syntactic and semantic information contained in the representations. In particular, subtracting these syntactic and semantic ``centroids'' from sentence vectors strongly affects their similarity with syntactically and semantically matched sentences, respectively, suggesting that syntax and semantics are, at least partially, linearly encoded. We also find that the cross-layer encoding profiles of syntax and semantics are different, and that the two signals can to some extent be decoupled, suggesting differential encoding of these two types of linguistic information in LLM representations.

Santiago Acevedo, Alex Rodriguez, Alessandro Laio

In real-world data, information is stored in extremely large feature vectors. These variables are typically correlated due to complex interactions involving many features simultaneously. Such correlations qualitatively correspond to semantic roles and are naturally recognized by both the human brain and artificial neural networks. This recognition enables, for instance, the prediction of missing parts of an image or text based on their context. We present a method to detect these correlations in high-dimensional data represented as binary numbers. We estimate the binary intrinsic dimension of a dataset, which quantifies the minimum number of independent coordinates needed to describe the data, and is therefore a proxy of semantic complexity. The proposed algorithm is largely insensitive to the so-called curse of dimensionality, and can therefore be used in big data analysis. We test this approach identifying phase transitions in model magnetic systems and we then apply it to the detection of semantic correlations of images and text inside deep neural networks.

Santiago Acevedo, Andrea Mascaretti, Riccardo Rende et al.

Deep neural networks are known to develop similar representations for semantically related data, even when they belong to different domains, such as an image and its description, or the same text in different languages. We present a method for quantitatively investigating this phenomenon by measuring the relative information content of the representations of semantically related data and probing how it is encoded into multiple tokens of large language models (LLMs) and vision transformers. Looking first at how LLMs process pairs of translated sentences, we identify inner ``semantic'' layers containing the most language-transferable information. We find moreover that, on these layers, a larger LLM (DeepSeek-V3) extracts significantly more general information than a smaller one (Llama3.1-8B). Semantic information of English text is spread across many tokens and it is characterized by long-distance correlations between tokens and by a causal left-to-right (i.e., past-future) asymmetry. We also identify layers encoding semantic information within visual transformers. We show that caption representations in the semantic layers of LLMs predict visual representations of the corresponding images. We observe significant and model-dependent information asymmetries between image and text representations.

Ali Hussaini Umar, Franky Kevin Nando Tezoh, Jean Barbier et al.

In supervised classification tasks, models are trained to predict a label for each data point. In real-world datasets, these labels are often noisy due to annotation errors. While the impact of label noise on the performance of deep learning models has been widely studied, its effects on the networks' hidden representations remain poorly understood. We address this gap by systematically comparing hidden representations using the Information Imbalance, a computationally efficient proxy of conditional mutual information. Through this analysis, we observe that the information content of the hidden representations follows a double descent as a function of the number of network parameters, akin to the behavior of the test error. We further demonstrate that in the underparameterized regime, representations learned with noisy labels are more informative than those learned with clean labels, while in the overparameterized regime, these representations are equally informative. Our results indicate that the representations of overparameterized networks are robust to label noise. We also found that the information imbalance between the penultimate and pre-softmax layers decreases with cross-entropy loss in the overparameterized regime. This offers a new perspective on understanding generalization in classification tasks. Extending our analysis to representations learned from random labels, we show that these perform worse than random features. This indicates that training on random labels drives networks much beyond lazy learning, as weights adapt to encode labels information.

Lorenzo Basile, Santiago Acevedo, Luca Bortolussi et al.

To gain insight into the mechanisms behind machine learning methods, it is crucial to establish connections among the features describing data points. However, these correlations often exhibit a high-dimensional and strongly nonlinear nature, which makes them challenging to detect using standard methods. This paper exploits the entanglement between intrinsic dimensionality and correlation to propose a metric that quantifies the (potentially nonlinear) correlation between high-dimensional manifolds. We first validate our method on synthetic data in controlled environments, showcasing its advantages and drawbacks compared to existing techniques. Subsequently, we extend our analysis to large-scale applications in neural network representations. Specifically, we focus on latent representations of multimodal data, uncovering clear correlations between paired visual and textual embeddings, whereas existing methods struggle significantly in detecting similarity. Our results indicate the presence of highly nonlinear correlation patterns between latent manifolds.