In-context learning (ICL) has emerged as a powerful paradigm for easily adapting Large Language Models (LLMs) to various tasks. However, our understanding of how ICL works remains limited. We explore a simple model of ICL in a controlled setup with synthetic training data to investigate ICL of univariate linear functions. We experiment with a range of GPT-2-like transformer models trained from scratch. Our findings challenge the prevailing narrative that transformers adopt algorithmic approaches like linear regression to learn a linear function in-context. These models fail to generalize beyond their training distribution, highlighting fundamental limitations in their capacity to infer abstract task structures. Our experiments lead us to propose a mathematically precise hypothesis of what the model might be learning.

Deep Neural Networks (DNNs) have revolutionized artificial intelligence, achieving impressive results on diverse data types, including images, videos, and texts. However, DNNs still lag behind Gradient Boosting Decision Trees (GBDT) on tabular data, a format extensively utilized across various domains. In this paper, we propose DOFEN, short for \textbf{D}eep \textbf{O}blivious \textbf{F}orest \textbf{EN}semble, a novel DNN architecture inspired by oblivious decision trees. DOFEN constructs relaxed oblivious decision trees (rODTs) by randomly combining conditions for each column and further enhances performance with a two-level rODT forest ensembling process. By employing this approach, DOFEN achieves state-of-the-art results among DNNs and further narrows the gap between DNNs and tree-based models on the well-recognized benchmark: Tabular Benchmark \citep{grinsztajn2022tree}, which includes 73 total datasets spanning a wide array of domains. The code of DOFEN is available at: \url{https://github.com/Sinopac-Digital-Technology-Division/DOFEN}.

Fréchet regression, an extension of classical linear regression to general metric spaces, offers a robust framework for modeling complex relationships between variables when the responses lie outside of Euclidean spaces. This approach is especially well suited to high-dimensional datasets, such as vector representations, with particular relevance to fields like imaging, where capturing nonlinear dependencies and the intrinsic data structure is critical for accurate modeling (Fréchet (1948), Petersen and Müller (2019), Bhattacharjee and Müller (2023), Qiu, Yu and Zhu (2024)). A significant consideration in Fréchet regression arises when predicting multiple responses simultaneously, as seen in multi-target or multidimensional problems (Zhang and Zhou (2007), Hyvönen, Jääsaari and Roos (2024)). Unlike traditional regression, where each observation corresponds to a single response, Fréchet regression can be extended to model complex interactions between multiple outputs. This ability to address complex relationships between several responses opens new avenues, particularly in fields such as bioinformatics (Huang et al. (2005)) and image analysis (Lathuilière et al. (2019)), where multidimensional data and interdependencies between responses require adaptive and specialized methodologies. However, to date, the handling of multilabel scenarios within the context of Fréchet regression remains relatively unexplored in the literature, despite its potential significance in addressing complex, multidimensional applications. In this paper, we present an extension of the Global Fréchet regression model, a specific variant of Fréchet regression that generalizes classical multiple linear regression by modeling responses as random objects. This extension enables the explicit modeling of relationships between input variables and multiple responses, thereby addressing the multi-label setting. Our second contribution in this paper addresses the dimensionality challenge in the context of the proposed Fréchet regression extension.

Gaussian graphical model selection is an important paradigm with numerous applications, including biological network modeling, financial network modeling, and social network analysis. Traditional approaches assume access to independent and identically distributed (i.i.d) samples, which is often impractical in real-world scenarios. In this paper, we address Gaussian graphical model selection under observations from a more realistic dependent stochastic process known as Glauber dynamics. Glauber dynamics, also called the Gibbs sampler, is a Markov chain that sequentially updates the variables of the underlying model based on the statistics of the remaining model. Such models, aside from frequently being employed to generate samples from complex multivariate distributions, naturally arise in various settings, such as opinion consensus in social networks and clearing/stock-price dynamics in financial networks. In contrast to the extensive body of existing work, we present the first algorithm for Gaussian graphical model selection when data are sampled according to the Glauber dynamics. We provide theoretical guarantees on the computational and statistical complexity of the proposed algorithm's structure learning performance. Additionally, we provide information-theoretic lower bounds on the statistical complexity and show that our algorithm is nearly minimax optimal for a broad class of problems.

This Dissertation is comprised of two main projects, addressing questions in neuroscience through applications of generative modeling. Project #1 (Chapter 4) explores how neurons encode features of the external world. I combine Helmholtz's "Perception as Unconscious Inference" -- paralleled by modern generative models like variational autoencoders (VAE) -- with the hierarchical structure of the visual cortex. This combination leads to the development of a hierarchical VAE model, which I test for its ability to mimic neurons from the primate visual cortex in response to motion stimuli. Results show that the hierarchical VAE perceives motion similar to the primate brain. Additionally, the model identifies causal factors of retinal motion inputs, such as object- and self-motion, in a completely unsupervised manner. Collectively, these results suggest that hierarchical inference underlines the brain's understanding of the world, and hierarchical VAEs can effectively model this understanding. Project #2 (Chapter 5) investigates the spatiotemporal structure of spontaneous brain activity and its reflection of brain states like rest. Using simultaneous fMRI and wide-field Ca2+ imaging data, this project demonstrates that the mouse cortex can be decomposed into overlapping communities, with around half of the cortical regions belonging to multiple communities. Comparisons reveal similarities and differences between networks inferred from fMRI and Ca2+ signals. The introduction (Chapter 1) is divided similarly to this abstract: sections 1.1 to 1.8 provide background information about Project #1, and sections 1.9 to 1.13 are related to Project #2. Chapter 2 includes historical background, Chapter 3 provides the necessary mathematical background, and finally, Chapter 6 contains concluding remarks and future directions.

This study explores the estimation of parameters in a matrix-valued linear regression model, where the $T$ responses $(Y_t)_{t=1}^T \in \mathbb{R}^{n \times p}$ and predictors $(X_t)_{t=1}^T \in \mathbb{R}^{m \times q}$ satisfy the relationship $Y_t = A^* X_t B^* + E_t$ for all $t = 1, \ldots, T$. In this model, $A^* \in \mathbb{R}_+^{n \times m}$ has $L_1$-normalized rows, $B^* \in \mathbb{R}^{q \times p}$, and $(E_t)_{t=1}^T$ are independent noise matrices following a matrix Gaussian distribution. The primary objective is to estimate the unknown parameters $A^*$ and $B^*$ efficiently. We propose explicit optimization-free estimators and establish non-asymptotic convergence rates to quantify their performance. Additionally, we extend our analysis to scenarios where $A^*$ and $B^*$ exhibit sparse structures. To support our theoretical findings, we conduct numerical simulations that confirm the behavior of the estimators, particularly with respect to the impact of the dimensions $n, m, p, q$, and the sample size $T$ on finite-sample performances. We complete the simulations by investigating the denoising performances of our estimators on noisy real-world images.

Group Re-identification (G-ReID) faces greater complexity than individual Re-identification (ReID) due to challenges like mutual occlusion, dynamic member interactions, and evolving group structures. Prior graph-based approaches have aimed to capture these dynamics by modeling the group as a single topological structure. However, these methods struggle to generalize across diverse group compositions, as they fail to fully represent the multifaceted relationships within the group. In this study, we introduce a Hierarchical Multi-Graphs Learning (HMGL) framework to address these challenges. Our approach models the group as a collection of multi-relational graphs, leveraging both explicit features (such as occlusion, appearance, and foreground information) and implicit dependencies between members. This hierarchical representation, encoded via a Multi-Graphs Neural Network (MGNN), allows us to resolve ambiguities in member relationships, particularly in complex, densely populated scenes. To further enhance matching accuracy, we propose a Multi-Scale Matching (MSM) algorithm, which mitigates issues of member information ambiguity and sensitivity to hard samples, improving robustness in challenging scenarios. Our method achieves state-of-the-art performance on two standard benchmarks, CSG and RoadGroup, with Rank-1/mAP scores of 95.3%/94.4% and 93.9%/95.4%, respectively. These results mark notable improvements of 1.7% and 2.5% in Rank-1 accuracy over existing approaches.

Parameter identifiability refers to the capability of accurately inferring the parameter values of a model from its observations (data). Traditional analysis methods exploit analytical properties of the closed form model, in particular sensitivity analysis, to quantify the response of the model predictions to variations in parameters. Techniques developed to analyze data, specifically manifold learning methods, have the potential to complement, and even extend the scope of the traditional analytical approaches. We report on a study comparing and contrasting analytical and data-driven approaches to quantify parameter identifiability and, importantly, perform parameter reduction tasks. We use the infinite bus synchronous generator model, a well-understood model from the power systems domain, as our benchmark problem. Our traditional analysis methods use the Fisher Information Matrix to quantify parameter identifiability analysis, and the Manifold Boundary Approximation Method to perform parameter reduction. We compare these results to those arrived at through data-driven manifold learning schemes: Output - Diffusion Maps and Geometric Harmonics. For our test case, we find that the two suites of tools (analytical when a model is explicitly available, as well as data-driven when the model is lacking and only measurement data are available) give (correct) comparable results; these results are also in agreement with traditional analysis based on singular perturbation theory. We then discuss the prospects of using data-driven methods for such model analysis.

Variational Autoencoders (VAE) are widely used for dimensionality reduction of large-scale tabular and image datasets, under the assumption of independence between data observations. In practice, however, datasets are often correlated, with typical sources of correlation including spatial, temporal and clustering structures. Inspired by the literature on linear mixed models (LMM), we propose LMMVAE -- a novel model which separates the classic VAE latent model into fixed and random parts. While the fixed part assumes the latent variables are independent as usual, the random part consists of latent variables which are correlated between similar clusters in the data such as nearby locations or successive measurements. The classic VAE architecture and loss are modified accordingly. LMMVAE is shown to improve squared reconstruction error and negative likelihood loss significantly on unseen data, with simulated as well as real datasets from various applications and correlation scenarios. It also shows improvement in the performance of downstream tasks such as supervised classification on the learned representations.

Generative methods (Gen-AI) are reviewed with a particular goal to solving tasks in Machine Learning and Bayesian inference. Generative models require one to simulate a large training dataset and to use deep neural networks to solve a supervised learning problem. To do this, we require high dimensional regression methods and tools for dimensionality reduction (a.k.a feature selection). The main advantage of Gen-AI methods is their ability to be model-free and to use deep neural networks to estimate conditional densities or posterior quantiles of interest. To illustrate generative methods, we analyze the well-known Ebola data-set. Finally, we conclude with directions for future research.