Transformer architectures have shown promising results in time series processing. However, despite recent advances in subquadratic attention mechanisms or state-space models, processing very long sequences still imposes significant computational requirements. Token merging, which involves replacing multiple tokens with a single one calculated as their linear combination, has shown to considerably improve the throughput of vision transformer architectures while maintaining accuracy. In this work, we go beyond computer vision and perform the first investigations of token merging in time series analysis on both time series transformers and state-space models. To effectively scale token merging to long sequences, we introduce local merging, a domain-specific token merging algorithm that selectively combines tokens within a local neighborhood, adjusting the computational complexity from linear to quadratic based on the neighborhood size. Our comprehensive empirical evaluation demonstrates that token merging offers substantial computational benefits with minimal impact on accuracy across various models and datasets. On the recently proposed Chronos foundation model, we achieve accelerations up to 5400 % with only minor accuracy degradations.

The advent of Transformers marked a significant breakthrough in sequence modelling, providing a highly performant architecture capable of leveraging GPU parallelism. However, Transformers are computationally expensive at inference time, limiting their applications, particularly in low-resource settings (e.g., mobile and embedded devices). Addressing this, we (1) begin by showing that attention can be viewed as a special Recurrent Neural Network (RNN) with the ability to compute its \textit{many-to-one} RNN output efficiently. We then (2) show that popular attention-based models such as Transformers can be viewed as RNN variants. However, unlike traditional RNNs (e.g., LSTMs), these models cannot be updated efficiently with new tokens, an important property in sequence modelling. Tackling this, we (3) introduce a new efficient method of computing attention's \textit{many-to-many} RNN output based on the parallel prefix scan algorithm. Building on the new attention formulation, we (4) introduce \textbf{Aaren}, an attention-based module that can not only (i) be trained in parallel (like Transformers) but also (ii) be updated efficiently with new tokens, requiring only constant memory for inferences (like traditional RNNs). Empirically, we show Aarens achieve comparable performance to Transformers on $38$ datasets spread across four popular sequential problem settings: reinforcement learning, event forecasting, time series classification, and time series forecasting tasks while being more time and memory-efficient.

Recent studies have highlighted their proficiency in some simple tasks like writing and coding through various reasoning strategies. However, LLM agents still struggle with tasks that require comprehensive planning, a process that challenges current models and remains a critical research issue. In this study, we concentrate on travel planning, a Multi-Phases planning problem, that involves multiple interconnected stages, such as outlining, information gathering, and planning, often characterized by the need to manage various constraints and uncertainties. Existing reasoning approaches have struggled to effectively address this complex task. Our research aims to address this challenge by developing a human-like planning framework for LLM agents, i.e., guiding the LLM agent to simulate various steps that humans take when solving Multi-Phases problems. Specifically, we implement several strategies to enable LLM agents to generate a coherent outline for each travel query, mirroring human planning patterns. Additionally, we integrate Strategy Block and Knowledge Block into our framework: Strategy Block facilitates information collection, while Knowledge Block provides essential information for detailed planning. Through our extensive experiments, we demonstrate that our framework significantly improves the planning capabilities of LLM agents, enabling them to tackle the travel planning task with improved efficiency and effectiveness. Our experimental results showcase the exceptional performance of the proposed framework; when combined with GPT-4-Turbo, it attains $10\times$ the performance gains in comparison to the baseline framework deployed on GPT-4-Turbo.

Recent studies have attempted to refine the Transformer architecture to demonstrate its effectiveness in Long-Term Time Series Forecasting (LTSF) tasks. Despite surpassing many linear forecasting models with ever-improving performance, we remain skeptical of Transformers as a solution for LTSF. We attribute the effectiveness of these models largely to the adopted Patch mechanism, which enhances sequence locality to an extent yet fails to fully address the loss of temporal information inherent to the permutation-invariant self-attention mechanism. Further investigation suggests that simple linear layers augmented with the Patch mechanism may outperform complex Transformer-based LTSF models. Moreover, diverging from models that use channel independence, our research underscores the importance of cross-variable interactions in enhancing the performance of multivariate time series forecasting. The interaction information between variables is highly valuable but has been misapplied in past studies, leading to suboptimal cross-variable models. Based on these insights, we propose a novel and simple Patch-based Decomposed MLP (PDMLP) for LTSF tasks. Specifically, we employ simple moving averages to extract smooth components and noise-containing residuals from time series data, engaging in semantic information interchange through channel mixing and specializing in random noise with channel independence processing. The PDMLP model consistently achieves state-of-the-art results on several real-world datasets. We hope this surprising finding will spur new research directions in the LTSF field and pave the way for more efficient and concise solutions.

Scaling law that rewards large datasets, complex models and enhanced data granularity has been observed in various fields of deep learning. Yet, studies on time series forecasting have cast doubt on scaling behaviors of deep learning methods for time series forecasting: while more training data improves performance, more capable models do not always outperform less capable models, and longer input horizons may hurt performance for some models. We propose a theory for scaling law for time series forecasting that can explain these seemingly abnormal behaviors. We take into account the impact of dataset size and model complexity, as well as time series data granularity, particularly focusing on the look-back horizon, an aspect that has been unexplored in previous theories. Furthermore, we empirically evaluate various models using a diverse set of time series forecasting datasets, which (1) verifies the validity of scaling law on dataset size and model complexity within the realm of time series forecasting, and (2) validates our theoretical framework, particularly regarding the influence of look back horizon. We hope our findings may inspire new models targeting time series forecasting datasets of limited size, as well as large foundational datasets and models for time series forecasting in future works.\footnote{Codes for our experiments will be made public at: \url{https://github.com/JingzheShi/ScalingLawForTimeSeriesForecasting}.

Ocean dynamics plays a crucial role in driving global weather and climate patterns. Accurate and efficient modeling of ocean dynamics is essential for improved understanding of complex ocean circulation and processes, for predicting climate variations and their associated teleconnections, and for addressing the challenges of climate change. While great efforts have been made to improve numerical Ocean General Circulation Models (OGCMs), accurate forecasting of global oceanic variations for multi-year remains to be a long-standing challenge. Here, we introduce ORCA (Oceanic Reliable foreCAst), the first data-driven model predicting global ocean circulation from multi-year to decadal time scales. ORCA accurately simulates the three-dimensional circulations and dynamics of the global ocean with high physical consistency. Hindcasts of key oceanic variables demonstrate ORCA's remarkable prediction skills in predicting ocean variations compared with state-of-the-art numerical OGCMs and abilities in capturing occurrences of extreme events at the subsurface ocean and ENSO vertical patterns. These results demonstrate the potential of data-driven ocean models for providing cheap, efficient, and accurate global ocean modeling and prediction. Moreover, ORCA stably and faithfully emulates ocean dynamics at decadal timescales, demonstrating its potential even for climate projections. The model will be available at https://github.com/OpenEarthLab/ORCA.

With the increasing collection of time series data from various domains, there arises a strong demand for general time series forecasting models pre-trained on a large number of time-series datasets to support a variety of downstream prediction tasks. Enabling general time series forecasting faces two challenges: how to obtain unified representations from multi-domian time series data, and how to capture domain-specific features from time series data across various domains for adaptive transfer in downstream tasks. To address these challenges, we propose a Register Assisted General Time Series Forecasting Model with Decomposed Frequency Learning (ROSE), a novel pre-trained model for time series forecasting. ROSE employs Decomposed Frequency Learning for the pre-training task, which decomposes coupled semantic and periodic information in time series with frequency-based masking and reconstruction to obtain unified representations across domains. We also equip ROSE with a Time Series Register, which learns to generate a register codebook to capture domain-specific representations during pre-training and enhances domain-adaptive transfer by selecting related register tokens on downstream tasks. After pre-training on large-scale time series data, ROSE achieves state-of-the-art forecasting performance on 8 real-world benchmarks. Remarkably, even in few-shot scenarios, it demonstrates competitive or superior performance compared to existing methods trained with full data.

Recent Transformer-based large language models (LLMs) demonstrate in-context learning ability to perform various functions based solely on the provided context, without updating model parameters. To fully utilize the in-context capabilities in time series forecasting (TSF) problems, unlike previous Transformer-based or LLM-based time series forecasting methods, we reformulate "time series forecasting tasks" as input tokens by constructing a series of (lookback, future) pairs within the tokens. This method aligns more closely with the inherent in-context mechanisms, and is more parameter-efficient without the need of using pre-trained LLM parameters. Furthermore, it addresses issues such as overfitting in existing Transformer-based TSF models, consistently achieving better performance across full-data, few-shot, and zero-shot settings compared to previous architectures.

Human mobility significantly impacts various aspects of society, including transportation, urban planning, and public health. The increasing availability of diverse mobility data and advancements in deep learning have revolutionized mobility modeling. Existing deep learning models, however, mainly study spatio-temporal patterns using trajectories and often fall short in capturing the underlying semantic interdependency among activities. Moreover, they are also constrained by the data source. These two factors thereby limit their realism and adaptability, respectively. Meanwhile, traditional activity-based models (ABMs) in transportation modeling rely on rigid assumptions and are costly and time-consuming to calibrate, making them difficult to adapt and scale to new regions, especially those regions with limited amount of required conventional travel data. To address these limitations, we develop a novel generative deep learning approach for human mobility modeling and synthesis, using ubiquitous and open-source data. Additionally, the model can be fine-tuned with local data, enabling adaptable and accurate representations of mobility patterns across different regions. The model is evaluated on a nationwide dataset of the United States, where it demonstrates superior performance in generating activity chains that closely follow ground truth distributions. Further tests using state- or city-specific datasets from California, Washington, and Mexico City confirm its transferability. This innovative approach offers substantial potential to advance mobility modeling research, especially in generating human activity chains as input for downstream activity-based mobility simulation models and providing enhanced tools for urban planners and policymakers.

Mathematical reasoning is an important capability of large language models~(LLMs) for real-world applications. To enhance this capability, existing work either collects large-scale math-related texts for pre-training, or relies on stronger LLMs (\eg GPT-4) to synthesize massive math problems. Both types of work generally lead to large costs in training or synthesis. To reduce the cost, based on open-source available texts, we propose an efficient way that trains a small LLM for math problem synthesis, to efficiently generate sufficient high-quality pre-training data. To achieve it, we create a dataset using GPT-4 to distill its data synthesis capability into the small LLM. Concretely, we craft a set of prompts based on human education stages to guide GPT-4, to synthesize problems covering diverse math knowledge and difficulty levels. Besides, we adopt the gradient-based influence estimation method to select the most valuable math-related texts. The both are fed into GPT-4 for creating the knowledge distillation dataset to train the small LLM. We leverage it to synthesize 6 million math problems for pre-training our JiuZhang3.0 model, which only needs to invoke GPT-4 API 9.3k times and pre-train on 4.6B data. Experimental results have shown that JiuZhang3.0 achieves state-of-the-art performance on several mathematical reasoning datasets, under both natural language reasoning and tool manipulation settings. Our code and data will be publicly released in \url{https://github.com/RUCAIBox/JiuZhang3.0}.