Search

Random forest (RF) stands out as a highly favored machine learning approach for classification problems. The effectiveness of RF hinges on two key factors: the accuracy of individual trees and the diversity among them. In this study, we introduce a novel approach called heterogeneous RF (HRF), designed to enhance tree diversity in a meaningful way. This diversification is achieved by deliberately introducing heterogeneity during the tree construction. Specifically, features used for splitting near the root node of previous trees are assigned lower weights when constructing the feature sub-space of the subsequent trees. As a result, dominant features in the prior trees are less likely to be employed in the next iteration, leading to a more diverse set of splitting features at the nodes. Through simulation studies, it was confirmed that the HRF method effectively mitigates the selection bias of trees within the ensemble, increases the diversity of the ensemble, and demonstrates superior performance on datasets with fewer noise features. To assess the comparative performance of HRF against other widely adopted ensemble methods, we conducted tests on 52 datasets, comprising both real-world and synthetic data. HRF consistently outperformed other ensemble methods in terms of accuracy across the majority of datasets., Comment: 18 pages, 6 figures

Accurately aligning contextual representations in cross-lingual sentence embeddings is key for effective parallel data mining. A common strategy for achieving this alignment involves disentangling semantics and language in sentence embeddings derived from multilingual pre-trained models. However, we discover that current disentangled representation learning methods suffer from semantic leakage - a term we introduce to describe when a substantial amount of language-specific information is unintentionally leaked into semantic representations. This hinders the effective disentanglement of semantic and language representations, making it difficult to retrieve embeddings that distinctively represent the meaning of the sentence. To address this challenge, we propose a novel training objective, ORthogonAlity Constraint LEarning (ORACLE), tailored to enforce orthogonality between semantic and language embeddings. ORACLE builds upon two components: intra-class clustering and inter-class separation. Through experiments on cross-lingual retrieval and semantic textual similarity tasks, we demonstrate that training with the ORACLE objective effectively reduces semantic leakage and enhances semantic alignment within the embedding space., Comment: 18 pages, 16 figures

The speed of an exhaustive search can be measured by a cover time, which is defined as the time it takes a random searcher to visit every state in some target set. Cover times have been studied in both the physics and probability literatures, with most prior works focusing on a single searcher. In this paper, we prove an explicit formula for all the moments of the cover time for many searchers on an arbitrary discrete network. Our results show that such cover times depend only on properties of the network along the shortest paths to the most distant parts of the target. This mere local dependence contrasts with the well-known result that cover times for single searchers depend on global properties of the network. We illustrate our rigorous results by stochastic simulations., Comment: 9 pages, 5 figures

Cover times measure the speed of exhaustive searches which require the exploration of an entire spatial region(s). Applications include the immune system hunting pathogens, animals collecting food, robotic demining or cleaning, and computer search algorithms. Mathematically, a cover time is the first time a random searcher(s) comes within a specified "detection radius" of every point in the target region (often the entire spatial domain). Due to their many applications and their fundamental probabilistic importance, cover times have been extensively studied in the physics and probability literatures. This prior work has generally studied cover times of a single searcher with a vanishing detection radius or a large target region. This prior work has further claimed that cover times for multiple searchers can be estimated by a simple rescaling of the cover time of a single searcher. In this paper, we study cover times of many diffusive or subdiffusive searchers and show that prior estimates break down as the number of searchers grows. We prove a rather universal formula for all the moments of such cover times in the many searcher limit that depends only on (i) the searcher's characteristic (sub)diffusivity and (ii) a certain geodesic distance between the searcher starting location(s) and the farthest point in the target. This formula is otherwise independent of the detection radius, space dimension, target size, and domain size. We illustrate our results in several examples and compare them to detailed stochastic simulations., Comment: 19 pages, 4 figures

Foraging is crucial for animals to survive. Many species forage in groups, as individuals communicate to share information about the location of available resources. For example, eusocial foragers, such as honey bees and many ants, recruit members from their central hive or nest to a known foraging site. However, the optimal level of communication and recruitment depends on the overall group size, the distribution of available resources, and the extent of interference between multiple individuals attempting to forage from a site. In this paper, we develop a discrete-time Markov chain model of eusocial foragers, who communicate information with a certain probability. We compare the stochastic model and its corresponding infinite-population limit. We find that foraging efficiency tapers off when recruitment probability is too high -- a phenomenon that does not occur in the infinite-population model, even though it occurs for any finite population size. The marginal inefficiency at high recruitment probability increases as the population increases, similar to a boundary layer. In particular, we prove there is a significant gap between the foraging efficiency of finite and infinite population models in the extreme case of complete communication. We also analyze this phenomenon by approximating the stationary distribution of foragers over sites in terms of mean escape times from multiple quasi-steady states. We conclude that for any finite group of foragers, an individual who has found a resource should only sometimes recruit others to the same resource. We discuss the relationship between our analysis and multi-agent multi-arm bandit problems.

Intercellular signaling has an important role in organism development, but not all communication occurs using the same mechanism. Here, we analyze the energy efficiency of intercellular signaling by two canonical mechanisms: diffusion of signaling molecules and direct transport mediated by signaling cellular protrusions. We show that efficient contact formation for direct transport can be established by an optimal rate of projecting protrusions, which depends on the availability of information about the location of the target cell. The optimal projection rate also depends on how signaling molecules are transported along the protrusion, in particular the ratio of the energy cost for contact formation and molecule synthesis. Also, we compare the efficiency of the two signaling mechanisms, under various model parameters. We find that the direct transport is favored over the diffusion when transporting a large amount of signaling molecules. There is a critical number of signaling molecules at which the efficiency of the two mechanisms are the same. The critical number is small when the distance between cells is far, which helps explain why protrusion-based mechanisms are observed in long-range cellular communications.

We investigate Turing pattern formation in a stochastic and spatially discretized version of a reaction diffusion advection (RDA) equation, which was previously introduced to model synaptogenesis in \textit{C. elegans}. The model describes the interactions between a passively diffusing molecular species and an advecting species that switches between anterograde and retrograde motor-driven transport (bidirectional transport). Within the context of synaptogenesis, the diffusing molecules can be identified with the protein kinase CaMKII and the advecting molecules as glutamate receptors. The stochastic dynamics evolves according to an RDA master equation, in which advection and diffusion are both modeled as hopping reactions along a one-dimensional array of chemical compartments. Carrying out a linear noise approximation of the RDA master equation leads to an effective Langevin equation, whose power spectrum provides a means of extending the definition of a Turing instability to stochastic systems, namely, in terms of the existence of a peak in the power spectrum at a non-zero spatial frequency. We thus show how noise can significantly extend the range over which spontaneous patterns occur, which is consistent with previous studies of RD systems., Comment: 26 pages, 8 figures