Multi-modal large language models (MLLMs) have enabled numerous advances in understanding and reasoning in domains like vision, but we have not yet seen this broad success for time-series. Although prior works on time-series MLLMs have shown promising performance in time-series forecasting, very few works show how an LLM could be used for time-series reasoning in natural language. We propose a novel multi-modal time-series LLM approach that learns generalizable information across various domains with powerful zero-shot performance. First, we train a lightweight time-series encoder on top of an LLM to directly extract time-series information. Then, we fine-tune our model with chain-of-thought augmented time-series tasks to encourage the model to generate reasoning paths. We show that our model learns a latent representation that reflects specific time-series features (e.g. slope, frequency), as well as outperforming GPT-4o on a set of zero-shot reasoning tasks on a variety of domains.

Dynamic Algorithm Configuration (DAC) addresses the challenge of dynamically setting hyperparameters of an algorithm for a diverse set of instances rather than focusing solely on individual tasks. Agents trained with Deep Reinforcement Learning (RL) offer a pathway to solve such settings. However, the limited generalization performance of these agents has significantly hindered the application in DAC. Our hypothesis is that a potential bias in the training instances limits generalization capabilities. We take a step towards mitigating this by selecting a representative subset of training instances to overcome overrepresentation and then retraining the agent on this subset to improve its generalization performance. For constructing the meta-features for the subset selection, we particularly account for the dynamic nature of the RL agent by computing time series features on trajectories of actions and rewards generated by the agent's interaction with the environment. Through empirical evaluations on the Sigmoid and CMA-ES benchmarks from the standard benchmark library for DAC, called DACBench, we discuss the potentials of our selection technique compared to training on the entire instance set. Our results highlight the efficacy of instance selection in refining DAC policies for diverse instance spaces.

Non-stationary systems are found throughout the world, from climate patterns under the influence of variation in carbon dioxide concentration, to brain dynamics driven by ascending neuromodulation. Accordingly, there is a need for methods to analyze non-stationary processes, and yet most time-series analysis methods that are used in practice, on important problems across science and industry, make the simplifying assumption of stationarity. One important problem in the analysis of non-stationary systems is the problem class that we refer to as Parameter Inference from a Non-stationary Unknown Process (PINUP). Given an observed time series, this involves inferring the parameters that drive non-stationarity of the time series, without requiring knowledge or inference of a mathematical model of the underlying system. Here we review and unify a diverse literature of algorithms for PINUP. We formulate the problem, and categorize the various algorithmic contributions. This synthesis will allow researchers to identify gaps in the literature and will enable systematic comparisons of different methods. We also demonstrate that the most common systems that existing methods are tested on - notably the non-stationary Lorenz process and logistic map - are surprisingly easy to perform well on using simple statistical features like windowed mean and variance, undermining the practice of using good performance on these systems as evidence of algorithmic performance. We then identify more challenging problems that many existing methods perform poorly on and which can be used to drive methodological advances in the field. Our results unify disjoint scientific contributions to analyzing non-stationary systems and suggest new directions for progress on the PINUP problem and the broader study of non-stationary phenomena.

Machine-learning approaches to algorithm-selection typically take data describing an instance as input. Input data can take the form of features derived from the instance description or fitness landscape, or can be a direct representation of the instance itself, i.e. an image or textual description. Regardless of the choice of input, there is an implicit assumption that instances that are similar will elicit similar performance from algorithm, and that a model is capable of learning this relationship. We argue that viewing algorithm-selection purely from an instance perspective can be misleading as it fails to account for how an algorithm `views' similarity between instances. We propose a novel `algorithm-centric' method for describing instances that can be used to train models for algorithm-selection: specifically, we use short probing trajectories calculated by applying a solver to an instance for a very short period of time. The approach is demonstrated to be promising, providing comparable or better results to computationally expensive landscape-based feature-based approaches. Furthermore, projecting the trajectories into a 2-dimensional space illustrates that functions that are similar from an algorithm-perspective do not necessarily correspond to the accepted categorisation of these functions from a human perspective.

More and more stock trading strategies are constructed using deep reinforcement learning (DRL) algorithms, but DRL methods originally widely used in the gaming community are not directly adaptable to financial data with low signal-to-noise ratios and unevenness, and thus suffer from performance shortcomings. In this paper, to capture the hidden information, we propose a DRL based stock trading system using cascaded LSTM, which first uses LSTM to extract the time-series features from stock daily data, and then the features extracted are fed to the agent for training, while the strategy functions in reinforcement learning also use another LSTM for training. Experiments in DJI in the US market and SSE50 in the Chinese stock market show that our model outperforms previous baseline models in terms of cumulative returns and Sharp ratio, and this advantage is more significant in the Chinese stock market, a merging market. It indicates that our proposed method is a promising way to build a automated stock trading system.

Time series are measured and analyzed across the sciences. One method of quantifying the structure of time series is by calculating a set of summary statistics or `features', and then representing a time series in terms of its properties as a feature vector. The resulting feature space is interpretable and informative, and enables conventional statistical learning approaches, including clustering, regression, and classification, to be applied to time-series datasets. Many open-source software packages for computing sets of time-series features exist across multiple programming languages, including catch22 (22 features: Matlab, R, Python, Julia), feasts (42 features: R), tsfeatures (63 features: R), Kats (40 features: Python), tsfresh (779 features: Python), and TSFEL (390 features: Python). However, there are several issues: (i) a singular access point to these packages is not currently available; (ii) to access all feature sets, users must be fluent in multiple languages; and (iii) these feature-extraction packages lack extensive accompanying methodological pipelines for performing feature-based time-series analysis, such as applications to time-series classification. Here we introduce a solution to these issues in an R software package called theft: Tools for Handling Extraction of Features from Time series. theft is a unified and extendable framework for computing features from the six open-source time-series feature sets listed above. It also includes a suite of functions for processing and interpreting the performance of extracted features, including extensive data-visualization templates, low-dimensional projections, and time-series classification operations. With an increasing volume and complexity of time-series datasets in the sciences and industry, theft provides a standardized framework for comprehensively quantifying and interpreting informative structure in time series.

The ubiquitous availability of wearable sensors is responsible for driving the Internet-of-Things but is also making an impact on sport sciences and precision medicine. While human activity recognition from smartphone data or other types of inertial measurement units (IMU) has evolved to one of the most prominent daily life examples of machine learning, the underlying process of time-series feature engineering still seems to be time-consuming. This lengthy process inhibits the development of IMU-based machine learning applications in sport science and precision medicine. This contribution discusses a feature engineering workflow, which automates the extraction of time-series feature on based on the FRESH algorithm (FeatuRe Extraction based on Scalable Hypothesis tests) to identify statistically significant features from synchronized IMU sensors (IMeasureU Ltd, NZ). The feature engineering workflow has five main steps: time-series engineering, automated time-series feature extraction, optimized feature extraction, fitting of a specialized classifier, and deployment of optimized machine learning pipeline. The workflow is discussed for the case of a user-specific running-walking classification, and the generalization to a multi-user multi-activity classification is demonstrated.

Recently, reinforcement learning models have achieved great success, completing complex tasks such as mastering Go and other games with higher scores than human players. Many of these models collect considerable data on the tasks and improve accuracy by extracting visual and time-series features using convolutional neural networks (CNNs) and recurrent neural networks, respectively. However, these networks have very high computational costs because they need to be trained by repeatedly using a large volume of past playing data. In this study, we propose a novel practical approach called reinforcement learning with convolutional reservoir computing (RCRC) model. The RCRC model has several desirable features: 1. it can extract visual and time-series features very fast because it uses random fixed-weight CNN and the reservoir computing model; 2. it does not require the training data to be stored because it extracts features without training and decides action with evolution strategy. Furthermore, the model achieves state of the art score in the popular reinforcement learning task. Incredibly, we find the random weight-fixed simple networks like only one dense layer network can also reach high score in the RL task.

Current clinical practice to monitor patients' health follows either regular or heuristic-based lab test (e.g. blood test) scheduling. Such practice not only gives rise to redundant measurements accruing cost, but may even lead to unnecessary patient discomfort. From the computational perspective, heuristic-based test scheduling might lead to reduced accuracy of clinical forecasting models. Computationally learning an optimal clinical test scheduling and measurement collection, is likely to lead to both, better predictive models and patient outcome improvement. We address the scheduling problem using deep reinforcement learning (RL) to achieve high predictive gain and low measurement cost, by scheduling fewer, but strategically timed tests. We first show that in the simulation our policy outperforms heuristic-based measurement scheduling with higher predictive gain or lower cost measured by accumulated reward. We then learn a scheduling policy for mortality forecasting in the real-world clinical dataset (MIMIC3), our learned policy is able to provide useful clinical insights. To our knowledge, this is the first RL application on multi-measurement scheduling problem in the clinical setting.