Decision trees are one of the most popular classifiers in the machine learning literature. While the most common decision tree learning algorithms treat data as a batch, numerous algorithms have been proposed to construct decision trees from a data stream. A standard training strategy involves augmenting the current tree by changing a leaf node into a split. Here we typically maintain counters in each leaf which allow us to determine the optimal split, and whether the split should be done. In this paper we focus on how to speed up the search for the optimal split when dealing with sparse binary features and a binary class. We focus on finding splits that have the approximately optimal information gain or Gini index. In both cases finding the optimal split can be done in $O(d)$ time, where $d$ is the number of features. We propose an algorithm that yields $(1 + α)$ approximation when using conditional entropy in amortized $O(α^{-1}(1 + m\log d) \log \log n)$ time, where $m$ is the number of 1s in a data point, and $n$ is the number of data points. Similarly, for Gini index, we achieve $(1 + α)$ approximation in amortized $O(α^{-1} + m \log d)$ time. Our approach is beneficial for sp
Ptychographic reconstructions in reflection geometries are commonly interpreted with the same two-dimensional thin-sample model used in transmission, yet the validity of this approximation has not been established. We develop a three-dimensional weak-scattering description of reflection ptychography and derive explicit thickness criteria for when a two-dimensional model remains accurate. Because the sampled axial spatial frequency range is dominated by the rotation of the Ewald sphere rather than its curvature, reflection geometries impose far stricter thin-sample conditions than transmission geometries. The allowable thickness is reduced by one to two orders of magnitude for a representative extreme ultraviolet geometry, depending on the tolerance for appearance of artifacts. Simulations verify that conventional two-dimensional reconstructions may exhibit the thickness-dependent artifacts as predicted by the theory, with particularly strong distortions near specular Bragg minima. We further show that incorporating the correct depth-dependent propagation into the forward model resolves these distortions and enables recovery of sample thickness. These results establish practical val
We show that every graph $G$ excluding an apex-forest $H$ as a minor has layered pathwidth at most $|V(H)|-2$, and that every graph $G$ excluding an apex-linear forest (such as a fan) $H$ as a minor has layered treedepth at most $|V(H)|-2$. We further show that both bounds are optimal. These results improve on recent results of Hodor, La, Micek, and Rambaud (2025): The first result improves the previous best-known bound by a multiplicative factor of $2$, while the second strengthens a previous quadratic bound. In addition, we reduce from quadratic to linear the bound on the $S$-focused treedepth $\mathrm{td}(G,S)$ for graphs $G$ with a prescribed set of vertices $S$ excluding models of paths in which every branch set intersects~$S$.
We provide proofs certifying that the structure theorem for vertex sets of bounded bidimensionality holds with polynomial bounds. The bidimensionality of vertex sets is a common generalisation of both treewidth and the face-cover-number of vertex sets in planar graphs. As such, it plays a crucial role in extensions of Courcelle's Theorem to $H$-minor-free graphs. Recently, bidimensionality and similar parameters have emerged as key for extensions of known parameterized algorithms for problems defined on a terminal set $R$. A prominent example for such a problem is Steiner Tree, which admits efficient algorithms on planar graphs whenever $R$ can be covered with few faces. Key to the algorithmic applications of bidimensionality is a structure theorem that explains how a graph $G$ can be decomposed into pieces where the behaviour of $R$ is highly controlled. One may see this structure theorem as a rooted analogue of Robertson and Seymour's celebrated Grid Theorem. Combining recent advances in obtaining polynomial bounds in the Graph Minors framework with new techniques for handling annotated vertex sets, we show that all parameters in the structure theorem above admit polynomial bound
The apparent nonunitary evolution in the black hole information paradox and recent work on describing wavefunction collapse via nonunitary nonlinear stochastic operators has motivated us to analyze whether mixed states can be distinguished from quickly oscillating pure states. We have demonstrated that the answer is no for all practical purposes if only strong nonpostselected measurements are performed. However, if weak measurements in postselected systems are used, mixed states and quickly oscillating states produce different results. An experimental procedure is proposed which could in principle determine the nature of mixed states stemming from blackbody radiation, decoherence, thermalization in solid state materials, Unruh radiation and Hawking radiation, among others. The analysis in this work applies to all fast oscillations, including those at Planck scale. As such, tabletop weak measurements can be used to probe (very specific) potential high energy behavior, where strong nonpostselected measurements cannot be applied. This work also demonstrates that weak measurements are not equivalent to a set of strong measurements without postselection since measurements which are impo
Industry practitioners are increasingly concerned with software that contains vulnerable versions of third-party dependencies that are included both directly and transitively. To address this problem, projects are encouraged to both (a)~quickly update to non-vulnerable versions of dependencies and (b)~be mindful of the update practices of the dependencies they choose to use. To this end, researchers have proposed metrics to measure the responsiveness of the development teams of the packages in keeping their dependencies updated: Mean-Time-To-Update (MTTU) and Mean-Time-To-Remediate (MTTR). While MTTU covers all dependencies, MTTR quantifies the time needed for a package to update its vulnerable dependencies. However, existing metrics fail to capture important nuances, such as considering floating versions and prioritizing recent updates, leading to inaccurate reflections of a development team's update practices. \textit{The goal of this study is to aid practitioners in understanding how quickly packages update their dependencies.} We propose two novel metrics, Mean-Time-To-Update for dependencies (MTTU) and Mean-Time-To-Remediate for vulnerable dependencies (MTTR), that overcome th
We study the problem of constructing simulations of a given randomized search algorithm \texttt{alg} with expected running time $O( \mathcal{O} \log \mathcal{O})$, where $\mathcal{O}$ is the optimal expected running time of any such simulation. Counterintuitively, these simulators can be dramatically faster than the original algorithm in getting alg to perform a single successful run, and this is done without any knowledge about alg, its running time distribution, etc. For example, consider an algorithm that randomly picks some integer $t$ according to some distribution over the integers, and runs for $t$ seconds. then with probability $1/2$ it stops, or else runs forever (i.e., a catastrophe). The simulators described here, for this case, all terminate in constant expected time, with exponentially decaying distribution on the running time of the simulation. Luby et al. studied this problem before -- and our main contribution is in offering several additional simulation strategies to the one they describe. In particular, one of our (optimal) simulation strategies is strikingly simple: Randomly pick an integer $t>0$ with probability $c/t^2$ (with $c= 6/π^2$). Run the algorithm fo
Foundation models like SAM (Segment Anything Model) exhibit strong zero-shot image segmentation performance, but often fall short on domain-specific tasks. Fine-tuning these models typically requires significant manual effort and domain expertise. In this work, we introduce QTT-SEG, a meta-learning-driven approach for automating and accelerating the fine-tuning of SAM for image segmentation. Built on the Quick-Tune hyperparameter optimization framework, QTT-SEG predicts high-performing configurations using meta-learned cost and performance models, efficiently navigating a search space of over 200 million possibilities. We evaluate QTT-SEG on eight binary and five multiclass segmentation datasets under tight time constraints. Our results show that QTT-SEG consistently improves upon SAM's zero-shot performance and surpasses AutoGluon Multimodal, a strong AutoML baseline, on most binary tasks within three minutes. On multiclass datasets, QTT-SEG delivers consistent gains as well. These findings highlight the promise of meta-learning in automating model adaptation for specialized segmentation tasks. Code available at: https://github.com/ds-brx/QTT-SEG/
Dimensionality is an important aspect for analyzing and understanding (high-dimensional) data. In their 2006 ICDM paper Tatti et al. answered the question for a (interpretable) dimension of binary data tables by introducing a normalized correlation dimension. In the present work we revisit their results and contrast them with a concept based notion of intrinsic dimension (ID) recently introduced for geometric data sets. To do this, we present a novel approximation for this ID that is based on computing concepts only up to a certain support value. We demonstrate and evaluate our approximation using all available datasets from Tatti et al., which have between 469 and 41271 extrinsic dimensions.
We give a short proof that for every apex-forest $X$ on at least two vertices, graphs excluding $X$ as a minor have layered pathwidth at most $2|V(X)|-3$. This improves upon a result by Dujmović, Eppstein, Joret, Morin, and Wood (SIDMA, 2020). Our main tool is a structural result about graphs excluding a forest as a rooted minor, which is of independent interest. We develop similar tools for treedepth and treewidth. We discuss implications for Erdős-Pósa properties of rooted models of minors in graphs.
With the ever-increasing number of pretrained models, machine learning practitioners are continuously faced with which pretrained model to use, and how to finetune it for a new dataset. In this paper, we propose a methodology that jointly searches for the optimal pretrained model and the hyperparameters for finetuning it. Our method transfers knowledge about the performance of many pretrained models with multiple hyperparameter configurations on a series of datasets. To this aim, we evaluated over 20k hyperparameter configurations for finetuning 24 pretrained image classification models on 87 datasets to generate a large-scale meta-dataset. We meta-learn a multi-fidelity performance predictor on the learning curves of this meta-dataset and use it for fast hyperparameter optimization on new datasets. We empirically demonstrate that our resulting approach can quickly select an accurate pretrained model for a new dataset together with its optimal hyperparameters.
The development of detectors with a high time resolution has been pivotal to our comprehension of neutron stars and the accurate measurement of their properties. While high-time resolution astronomy has become a standard in the radio and the high-/very-high-energy bands, progress in the visible band has been comparatively much slower. SiFAP2 is a high-speed optical photometer mounted at the INAF Telescopio Nazionale Galileo. Its potential emerged with the discovery of the first two optical millisecond pulsars: these are among the most efficient particle accelerators and natural laboratories of fundamental physics. Optical millisecond pulsations challenge the standard pulsar paradigm, requiring innovative solutions. Higher photon counting statistics of optical telescopes, compared to high-energy instruments, attain unprecedented sensitivity for weak pulsed signals from bright accreting neutron stars, which are the best candidates for still undetected continuous gravitational waves.
Individual human recognition is important for species that live in close proximity to humans. Numerous studies on domesticated species and urban-adapted birds have highlighted this ability. One such species which is heavily reliant on humans is the free-ranging dog. Very little knowledge exists on the amount of time taken by free-ranging dogs to learn and remember individual humans. Due to their territorial nature, they have a high probability of encountering the same people multiple times on the streets. Being able to distinguish individual humans might be helpful in making decisions regarding people from whom to beg for food or social reward. We investigated if free-ranging dogs are capable of identifying the person rewarding them and the amount of time required for them to learn it. We conducted field trials on randomly selected adult free-ranging dogs in West Bengal, India. On Day 1, a choice test was conducted. The experimenter chosen did not provide reward while the other experimenter provided a piece of boiled chicken followed by petting. The person giving reward on Day 1 served as the correct choice on four subsequent days of training. Day 6 was the test day when none of th
In place of in-house solutions, organizations are increasingly moving towards managed services for cyber defense. Security Operations Centers are specialized cybersecurity units responsible for the defense of an organization, but the large-scale centralization of threat detection is causing SOCs to endure an overwhelming amount of false positive alerts -- a phenomenon known as alert fatigue. Large collections of imprecise sensors, an inability to adapt to known false positives, evolution of the threat landscape, and inefficient use of analyst time all contribute to the alert fatigue problem. To combat these issues, we present That Escalated Quickly (TEQ), a machine learning framework that reduces alert fatigue with minimal changes to SOC workflows by predicting alert-level and incident-level actionability. On real-world data, the system is able to reduce the time it takes to respond to actionable incidents by $22.9\%$, suppress $54\%$ of false positives with a $95.1\%$ detection rate, and reduce the number of alerts an analyst needs to investigate within singular incidents by $14\%$.
This paper considers elections in which voters choose one candidate each, independently according to known probability distributions. A candidate receiving a strict majority (absolute or relative, depending on the version) wins. After the voters have made their choices, each vote can be inspected to determine which candidate received that vote. The time (or cost) to inspect each of the votes is known in advance. The task is to (possibly adaptively) determine the order in which to inspect the votes, so as to minimize the expected time to determine which candidate has won the election. We design polynomial-time constant-factor approximation algorithms for both the absolute-majority and the relative-majority version. Both algorithms are based on a two-phase approach. In the first phase, the algorithms reduce the number of relevant candidates to $O(1)$, and in the second phase they utilize techniques from the literature on stochastic function evaluation to handle the remaining candidates. In the case of absolute majority, we show that the same can be achieved with only two rounds of adaptivity.
We consider the max-min fair resource allocation problem. The best-known solutions use either a sequence of optimizations or waterfilling, which only applies to a narrow set of cases. These solutions have become a practical bottleneck in WAN traffic engineering and cluster scheduling, especially at larger problem sizes. We improve both approaches: (1) we show how to convert the optimization sequence into a single fast optimization, and (2) we generalize waterfilling to the multi-path case. We empirically show our new algorithms Pareto-dominate prior techniques: they produce faster, fairer, and more efficient allocations. Some of our allocators also have theoretical guarantees: they trade off a bounded amount of unfairness for faster allocation. We have deployed our allocators in Azure's WAN traffic engineering pipeline, where we preserve solution quality and achieve a roughly $3\times$ speedup.
Manual of practical experiments on protocols used in the Internet or TCP/IP Suite. This manual is a collection of experiments that have been used in undergraduate and graduate courses taught at New Jersey Institute of Technology for a few years (ECE429 and ECE637). The manual is updated periodically to accommodate emerging needs and technologies, including virtualization of the experiments for their use during lockdowns as the one we experienced during the COVID-19 pandemic in 2020 and part of 2021. The manual may be used by all those interested in knowing and experiencing some of the basic protocols that run on the Internet. The goal is to get some understanding on protocol design and not necessarily on a particular software or operative system. Yet, the exercises will give the interested person a practical perspective. Experiments are targeted towards using Linux but most of them can also be performed on Windows operating systems.
Partial differential equations (PDEs) are often computationally challenging to solve, and in many settings many related PDEs must be be solved either at every timestep or for a variety of candidate boundary conditions, parameters, or geometric domains. We present a meta-learning based method which learns to rapidly solve problems from a distribution of related PDEs. We use meta-learning (MAML and LEAP) to identify initializations for a neural network representation of the PDE solution such that a residual of the PDE can be quickly minimized on a novel task. We apply our meta-solving approach to a nonlinear Poisson's equation, 1D Burgers' equation, and hyperelasticity equations with varying parameters, geometries, and boundary conditions. The resulting Meta-PDE method finds qualitatively accurate solutions to most problems within a few gradient steps; for the nonlinear Poisson and hyper-elasticity equation this results in an intermediate accuracy approximation up to an order of magnitude faster than a baseline finite element analysis (FEA) solver with equivalent accuracy. In comparison to other learned solvers and surrogate models, this meta-learning approach can be trained without
A genome read data set can be quickly and efficiently remapped from one reference to another similar reference (e.g., between two reference versions or two similar species) using a variety of tools, e.g., the commonly-used CrossMap tool. With the explosion of available genomic data sets and references, high-performance remapping tools will be even more important for keeping up with the computational demands of genome assembly and analysis. We provide FastRemap, a fast and efficient tool for remapping reads between genome assemblies. FastRemap provides up to a 7.82$\times$ speedup (6.47$\times$, on average) and uses as low as 61.7% (80.7%, on average) of the peak memory consumption compared to the state-of-the-art remapping tool, CrossMap. FastRemap is written in C++. The source code and user manual are freely available at: github.com/CMU-SAFARI/FastRemap. Docker image available at: https://hub.docker.com/r/alkanlab/fast. Also available in Bioconda at: https://anaconda.org/bioconda/fastremap-bio.
Multi-object manipulation problems in continuous state and action spaces can be solved by planners that search over sampled values for the continuous parameters of operators. The efficiency of these planners depends critically on the effectiveness of the samplers used, but effective sampling in turn depends on details of the robot, environment, and task. Our strategy is to learn functions called "specializers" that generate values for continuous operator parameters, given a state description and values for the discrete parameters. Rather than trying to learn a single specializer for each operator from large amounts of data on a single task, we take a modular meta-learning approach. We train on multiple tasks and learn a variety of specializers that, on a new task, can be quickly adapted using relatively little data -- thus, our system "learns quickly to plan quickly" using these specializers. We validate our approach experimentally in simulated 3D pick-and-place tasks with continuous state and action spaces. Visit http://tinyurl.com/chitnis-icra-19 for a supplementary video.