The rapid growth of Large Language Models (LLMs) has been a driving force in transforming various domains, reshaping the artificial general intelligence landscape. However, the increasing computational and memory demands of these models present substantial challenges, hindering both academic research and practical applications. To address these issues, a wide array of methods, including both algorithmic and hardware solutions, have been developed to enhance the efficiency of LLMs. This survey delivers a comprehensive review of algorithmic advancements aimed at improving LLM efficiency. Unlike other surveys that typically focus on specific areas such as training or model compression, this paper examines the multi-faceted dimensions of efficiency essential for the end-to-end algorithmic development of LLMs. Specifically, it covers various topics related to efficiency, including scaling laws, data utilization, architectural innovations, training and tuning strategies, and inference techniques. This paper aims to serve as a valuable resource for researchers and practitioners, laying the groundwork for future innovations in this critical research area. Our repository of relevant references is maintained at url{https://github.com/tding1/Efficient-LLM-Survey}.

The objective of this study is to evaluate whether the number of requisite training samples can be reduced with the use of various transfer learning models for predicting, for example, the chemical source terms of the data-driven reduced-order model that represents the homogeneous ignition process of a hydrogen/air mixture. Principal component analysis is applied to reduce the dimensionality of the hydrogen/air mixture in composition space. Artificial neural networks (ANNs) are used to tabulate the reaction rates of principal components, and subsequently, a system of ordinary differential equations is solved. As the number of training samples decreases at the target task (i.e.,for T0 > 1000 K and various phi), the reduced-order model fails to predict the ignition evolution of a hydrogen/air mixture. Three transfer learning strategies are then applied to the training of the ANN model with a sparse dataset. The performance of the reduced-order model with a sparse dataset is found to be remarkably enhanced if the training of the ANN model is restricted by a regularization term that controls the degree of knowledge transfer from source to target tasks. To this end, a novel transfer learning method is introduced, parameter control via partial initialization and regularization (PaPIR), whereby the amount of knowledge transferred is systemically adjusted for the initialization and regularization of the ANN model in the target task. It is found that an additional performance gain can be achieved by changing the initialization scheme of the ANN model in the target task when the task similarity between source and target tasks is relatively low.

Recent advances in deep learning have significantly elevated weather prediction models. However, these models often falter in real-world scenarios due to their sensitivity to spatial-temporal shifts. This issue is particularly acute in weather forecasting, where models are prone to overfit to local and temporal variations, especially when tasked with fine-grained predictions. In this paper, we address these challenges by developing a robust precipitation forecasting model that demonstrates resilience against such spatial-temporal discrepancies. We introduce Temporal Frame Interpolation (TFI), a novel technique that enhances the training dataset by generating synthetic samples through interpolating adjacent frames from satellite imagery and ground radar data, thus improving the model's robustness against frame noise. Moreover, we incorporate a unique Multi-Level Dice (ML-Dice) loss function, leveraging the ordinal nature of rainfall intensities to improve the model's performance. Our approach has led to significant improvements in forecasting precision, culminating in our model securing \textit{1st place} in the transfer learning leaderboard of the \textit{Weather4cast'23} competition. This achievement not only underscores the effectiveness of our methodologies but also establishes a new standard for deep learning applications in weather forecasting. Our code and weights have been public on \url{https://github.com/Secilia-Cxy/UNetTFI}.

The recent boom of linear forecasting models questions the ongoing passion for architectural modifications of Transformer-based forecasters. These forecasters leverage Transformers to model the global dependencies over temporal tokens of time series, with each token formed by multiple variates of the same timestamp. However, Transformers are challenged in forecasting series with larger lookback windows due to performance degradation and computation explosion. Besides, the embedding for each temporal token fuses multiple variates that represent potential delayed events and distinct physical measurements, which may fail in learning variate-centric representations and result in meaningless attention maps. In this work, we reflect on the competent duties of Transformer components and repurpose the Transformer architecture without any modification to the basic components. We propose iTransformer that simply applies the attention and feed-forward network on the inverted dimensions. Specifically, the time points of individual series are embedded into variate tokens which are utilized by the attention mechanism to capture multivariate correlations; meanwhile, the feed-forward network is applied for each variate token to learn nonlinear representations. The iTransformer model achieves state-of-the-art on challenging real-world datasets, which further empowers the Transformer family with promoted performance, generalization ability across different variates, and better utilization of arbitrary lookback windows, making it a nice alternative as the fundamental backbone of time series forecasting.

Most machine learning methods require tuning of hyper-parameters. For kernel ridge regression with the Gaussian kernel, the hyper-parameter is the bandwidth. The bandwidth specifies the length scale of the kernel and has to be carefully selected to obtain a model with good generalization. The default methods for bandwidth selection, cross-validation and marginal likelihood maximization, often yield good results, albeit at high computational costs. Inspired by Jacobian regularization, we formulate an approximate expression for how the derivatives of the functions inferred by kernel ridge regression with the Gaussian kernel depend on the kernel bandwidth. We use this expression to propose a closed-form, computationally feather-light, bandwidth selection heuristic, based on controlling the Jacobian. In addition, the Jacobian expression illuminates how the bandwidth selection is a trade-off between the smoothness of the inferred function and the conditioning of the training data kernel matrix. We show on real and synthetic data that compared to cross-validation and marginal likelihood maximization, our method is on pair in terms of model performance, but up to six orders of magnitude faster.

In the fall of 2018, the University of California tasked a team of researchers with identifying projects from which it could confidently purchase carbon offsets that would reliably cancel out greenhouse gas emissions across its campuses. They found next to nothing. The findings helped prompt the entire university system to radically rethink its sustainability plans. Now the researchers are sharing the lessons they learned over the course of the project, in the hopes of helping other universities and organizations consider what role, if any, offsets should play in sustainability strategies, MIT Technology Review can report. The project's leaders have three main takeaways for what others should do.

Forecasting natural gas consumption, considering seasonality and trends, is crucial in planning its supply and consumption and optimizing the cost of obtaining it, mainly by industrial entities. However, in times of threats to its supply, it is also a critical element that guarantees the supply of this raw material to meet individual consumers' needs, ensuring society's energy security. This article introduces a novel multistep ahead forecasting of natural gas consumption with change point detection integration for model collection selection with continual learning capabilities using data stream processing. The performance of the forecasting models based on the proposed approach is evaluated in a complex real-world use case of natural gas consumption forecasting. We employed Hoeffding tree predictors as forecasting models and the Pruned Exact Linear Time (PELT) algorithm for the change point detection procedure. The change point detection integration enables selecting a different model collection for successive time frames. Thus, three model collection selection procedures (with and without an error feedback loop) are defined and evaluated for forecasting scenarios with various densities of detected change points. These models were compared with change point agnostic baseline approaches. Our experiments show that fewer change points result in a lower forecasting error regardless of the model collection selection procedure employed. Also, simpler model collection selection procedures omitting forecasting error feedback leads to more robust forecasting models suitable for continual learning tasks.

Deep learning models, particularly Transformers, have achieved impressive results in various domains, including time series forecasting. While existing time series literature primarily focuses on model architecture modifications and data augmentation techniques, this paper explores the training schema of deep learning models for time series; how models are trained regardless of their architecture. We perform extensive experiments to investigate the occurrence of deep double descent in several Transformer models trained on public time series data sets. We demonstrate epoch-wise deep double descent and that overfitting can be reverted using more epochs. Leveraging these findings, we achieve state-of-the-art results for long sequence time series forecasting in nearly 70% of the 72 benchmarks tested. This suggests that many models in the literature may possess untapped potential. Additionally, we introduce a taxonomy for classifying training schema modifications, covering data augmentation, model inputs, model targets, time series per model, and computational budget.

Machine Learning (ML) and linear System Identification (SI) have been historically developed independently. In this paper, we leverage well-established ML tools - especially the automatic differentiation framework - to introduce SIMBa, a family of discrete linear multi-step-ahead state-space SI methods using backpropagation. SIMBa relies on a novel Linear-Matrix-Inequality-based free parametrization of Schur matrices to ensure the stability of the identified model. We show how SIMBa generally outperforms traditional linear state-space SI methods, and sometimes significantly, although at the price of a higher computational burden. This performance gap is particularly remarkable compared to other SI methods with stability guarantees, where the gain is frequently above 25% in our investigations, hinting at SIMBa's ability to simultaneously achieve state-of-the-art fitting performance and enforce stability. Interestingly, these observations hold for a wide variety of input-output systems and on both simulated and real-world data, showcasing the flexibility of the proposed approach. We postulate that this new SI paradigm presents a great extension potential to identify structured nonlinear models from data, and we hence open-source SIMBa on https://github.com/Cemempamoi/simba.

In recent years, there has been a growing interest in using machine learning (particularly deep learning) for modeling dynamical systems (Kocijan et al., 2004; Hedjar, 2013; Lenz et al., 2015; Moriyasu et al., 2019). Unlike the traditional modeling approach, machine learning can be used to create highly accurate models more easily without the need for detailed domain knowledge. However, control design using such models can become challenging because of the strong nonlinearity of the machine-learning models. Model predictive control (MPC) is a typical control method that can handle nonlinearity, wherein the model is used to optimize the predicted future behavior of the system; however, the optimal control problem solved each time is often nonconvex (Ławryńczuk, 2008; Nghiem, 2019; Gros, 2019) when the model is highly nonlinear, thereby making it difficult to ensure the uniqueness and optimality of the numerically obtained solution and the continuity of the control law (Moriyasu et al., 2022). Various nonlinear control theories besides MPC cannot be easily applied to general machine learning models because the model structures to which the theories can be applied are often limited to specific systems, such as input-affine systems.