We study the image-based geolocalization problem, aiming to localize ground-view query images on cartographic maps. Current methods often utilize cross-view localization techniques to match ground-view query images with 2D maps. However, the performance of these methods is unsatisfactory due to significant cross-view appearance differences. In this paper, we lift cross-view matching to a 2.5D space, where heights of structures (e.g., trees and buildings) provide geometric information to guide the cross-view matching. We propose a new approach to learning representative embeddings from multi-modal data. Specifically, we establish a projection relationship between 2.5D space and 2D aerial-view space. The projection is further used to combine multi-modal features from the 2.5D and 2D maps using an effective pixel-to-point fusion method. By encoding crucial geometric cues, our method learns discriminative location embeddings for matching panoramic images and maps. Additionally, we construct the first large-scale ground-to-2.5D map geolocalization dataset to validate our method and facilitate future research. Both single-image based and route based localization experiments are conducted to test our method. Extensive experiments demonstrate that the proposed method achieves significantly higher localization accuracy and faster convergence than previous 2D map-based approaches.

Decision Trees (DTs) are commonly used for many machine learning tasks due to their high degree of interpretability. However, learning a DT from data is a difficult optimization problem, as it is non-convex and non-differentiable. Therefore, common approaches learn DTs using a greedy growth algorithm that minimizes the impurity locally at each internal node. Unfortunately, this greedy procedure can lead to inaccurate trees. In this paper, we present a novel approach for learning hard, axis-aligned DTs with gradient descent. The proposed method uses backpropagation with a straight-through operator on a dense DT representation, to jointly optimize all tree parameters. Our approach outperforms existing methods on binary classification benchmarks and achieves competitive results for multi-class tasks. The method is available under: https://github.com/s-marton/GradTree

Standard probabilistic sparse coding assumes a Laplace prior, a linear mapping from latents to observables, and Gaussian observable distributions. We here derive a solely entropy-based learning objective for the parameters of standard sparse coding. The novel variational objective has the following features: (A) unlike MAP approximations, it uses non-trivial posterior approximations for probabilistic inference; (B) unlike for previous non-trivial approximations, the novel objective is fully analytical; and (C) the objective allows for a novel principled form of annealing. The objective is derived by first showing that the standard ELBO objective converges to a sum of entropies, which matches similar recent results for generative models with Gaussian priors. The conditions under which the ELBO becomes equal to entropies are then shown to have analytical solutions, which leads to the fully analytical objective. Numerical experiments are used to demonstrate the feasibility of learning with such entropy-based ELBOs. We investigate different posterior approximations including Gaussians with correlated latents and deep amortized approximations. Furthermore, we numerically investigate entropy-based annealing which results in improved learning. Our main contributions are theoretical, however, and they are twofold: (1) for non-trivial posterior approximations, we provide the (to the knowledge of the authors) first analytical ELBO objective for standard probabilistic sparse coding; and (2) we provide the first demonstration on how a recently shown convergence of the ELBO to entropy sums can be used for learning.

Classical analysis of convex and non-convex optimization methods often requires the Lipshitzness of the gradient, which limits the analysis to functions bounded by quadratics. Recent work relaxed this requirement to a non-uniform smoothness condition with the Hessian norm bounded by an affine function of the gradient norm, and proved convergence in the non-convex setting via gradient clipping, assuming bounded noise. In this paper, we further generalize this non-uniform smoothness condition and develop a simple, yet powerful analysis technique that bounds the gradients along the trajectory, thereby leading to stronger results for both convex and non-convex optimization problems. In particular, we obtain the classical convergence rates for (stochastic) gradient descent and Nesterov's accelerated gradient method in the convex and/or non-convex setting under this general smoothness condition. The new analysis approach does not require gradient clipping and allows heavy-tailed noise with bounded variance in the stochastic setting.

We consider the community recovery problem on a multilayer variant of the hypergraph stochastic block model (HSBM). Each layer is associated with an independent realization of a d-uniform HSBM on N vertices. Given the similarity matrix containing the aggregated number of hyperedges incident to each pair of vertices, the goal is to obtain a partition of the N vertices into disjoint communities. In this work, we investigate a semidefinite programming (SDP) approach and obtain information-theoretic conditions on the model parameters that guarantee exact recovery both in the assortative and the disassortative cases.

This study proposes a new approach that investigates differences in topological characteristics of visual networks, which are constructed using fMRI BOLD time-series corresponding to visual datasets of COCO, ImageNet, and SUN. A publicly available BOLD5000 dataset is utilized that contains fMRI scans while viewing 5254 images of diverse complexities. The objective of this study is to examine how network topology differs in response to distinct visual stimuli from these visual datasets. To achieve this, 0- and 1-dimensional persistence diagrams are computed for each visual network representing COCO, ImageNet, and SUN. For extracting suitable features from topological persistence diagrams, K-means clustering is executed. The extracted K-means cluster features are fed to a novel deep-hybrid model that yields accuracy in the range of 90%-95% in classifying these visual networks. To understand vision, this type of visual network categorization across visual datasets is important as it captures differences in BOLD signals while perceiving images with different contexts and complexities. Furthermore, distinctive topological patterns of visual network associated with each dataset, as revealed from this study, could potentially lead to the development of future neuroimaging biomarkers for diagnosing visual processing disorders like visual agnosia or prosopagnosia, and tracking changes in visual cognition over time.

Data attribution methods play a crucial role in understanding machine learning models, providing insight into which training data points are most responsible for model outputs during deployment. However, current state-of-the-art approaches require a large ensemble of as many as 300,000 models to accurately attribute model predictions. These approaches therefore come at a high computational cost, are memory intensive, and are hard to scale to large models or datasets. In this work, we focus on a minimalist baseline, utilizing the feature space of a backbone pretrained via self-supervised learning to perform data attribution. Our method is model-agnostic and scales easily to large datasets. We show results on CIFAR-10 and ImageNet, achieving strong performance that rivals or outperforms state-of-the-art approaches at a fraction of the compute or memory cost. Contrary to prior work, our results reinforce the intuition that a model's prediction on one image is most impacted by visually similar training samples. Our approach serves as a simple and efficient baseline for data attribution on images.

Fault prediction in time series data is a vital machine learning task with extensive industrial applications, yet it faces challenges such as data scarcity and frequency mismatch. Meta-learning has emerged as a promising approach to address these issues, leveraging cross-task similarities and differences to effectively adapt to novel time series fault prediction tasks. It empowers deep learning models to rapidly adjust to new time series data with few or even no samples, capitalizing on the similarities and differences among time series data from various domains and scenarios to enhance generalization capabilities([35], [2]). Meta-learning enables a machine learning algorithm to'learn to learn', enhancing the universality and adaptability of knowledge. In the realm of time series fault prediction, the efficacy of meta-learning hinges on the nuanced calibration of several task-distribution-dependent factors, of which researchers identify four key aspects: data representation, meta-learner design, meta-learning algorithms, and pseudo meta-task division. It's noteworthy that the first three aspects require different adjustments based on the specific task distribution, whereas the division of pseudo meta-tasks is not dependent on task distribution [24]. Therefore, to enhance the adaptability of meta-learning in fault prediction, this paper primarily refines the division method of pseudo meta-tasks.

As a result, many tasks that one would traditionally attribute to being done by a human being are being performed by machine learning (ML) and artificial intelligence (AI) based models. Hence it is not surprising to see machine learning models being deployed in areas where historically, due to the nature of the tasks, it would require the involvement of a human, e.g., getting a loan/ receiving a bail judgment. Unfortunately, many of these state-of-the-art machine learning or artificial intelligence-based systems are so complex that we are unable to understand why they made such a decision. This lack of clarity contributes to such models being viewed as a black box whose content/logic is unknown. A natural consequence of the increasing involvement of machine learning models in decisionmaking processes is that these decisions either directly or indirectly impact individuals.

Neurons can display highly variable dynamics. While such variability presumably supports the wide range of behaviors generated by the organism, their gene expressions are relatively stable in the adult brain. This suggests that neuronal activity is a combination of its time-invariant identity and the inputs the neuron receives from the rest of the circuit. Here, we propose a self-supervised learning based method to assign time-invariant representations to individual neurons based on permutation-, and population size-invariant summary of population recordings. We fit dynamical models to neuronal activity to learn a representation by considering the activity of both the individual and the neighboring population. Our self-supervised approach and use of implicit representations enable robust inference against imperfections such as partial overlap of neurons across sessions, trial-to-trial variability, and limited availability of molecular (transcriptomic) labels for downstream supervised tasks. We demonstrate our method on a public multimodal dataset of mouse cortical neuronal activity and transcriptomic labels. We report > 35% improvement in predicting the transcriptomic subclass identity and > 20% improvement in predicting class identity with respect to the state-of-the-art.