The emergence of Multimodal Large Language Models ((M)LLMs) has ushered in new avenues in artificial intelligence, particularly for autonomous driving by offering enhanced understanding and reasoning capabilities. This paper introduces LimSim++, an extended version of LimSim designed for the application of (M)LLMs in autonomous driving. Acknowledging the limitations of existing simulation platforms, LimSim++ addresses the need for a long-term closed-loop infrastructure supporting continuous learning and improved generalization in autonomous driving. The platform offers extended-duration, multi-scenario simulations, providing crucial information for (M)LLM-driven vehicles. Users can engage in prompt engineering, model evaluation, and framework enhancement, making LimSim++ a versatile tool for research and practice. This paper additionally introduces a baseline (M)LLM-driven framework, systematically validated through quantitative experiments across diverse scenarios. The open-source resources of LimSim++ are available at: https://pjlab-adg.github.io/limsim_plus/.

Low-resolution infrared (IR) array sensors enable people counting applications such as monitoring the occupancy of spaces and people flows while preserving privacy and minimizing energy consumption. Deep Neural Networks (DNNs) have been shown to be well-suited to process these sensor data in an accurate and efficient manner. Nevertheless, the space of DNNs' architectures is huge and its manual exploration is burdensome and often leads to sub-optimal solutions. To overcome this problem, in this work, we propose a highly automated full-stack optimization flow for DNNs that goes from neural architecture search, mixed-precision quantization, and post-processing, down to the realization of a new smart sensor prototype, including a Microcontroller with a customized instruction set. Integrating these cross-layer optimizations, we obtain a large set of Pareto-optimal solutions in the 3D-space of energy, memory, and accuracy. Deploying such solutions on our hardware platform, we improve the state-of-the-art achieving up to 4.2x model size reduction, 23.8x code size reduction, and 15.38x energy reduction at iso-accuracy.

Rain precipitation prediction is a challenging task as it depends on weather and meteorological features which vary from location to location. As a result, a prediction model that performs well at one location does not perform well at other locations due to the distribution shifts. In addition, due to global warming, the weather patterns are changing very rapidly year by year which creates the possibility of ineffectiveness of those models even at the same location as time passes. In our work, we have proposed an adaptive deep learning-based framework in order to provide a solution to the aforementioned challenges. Our method can generalize the model for the prediction of precipitation for any location where the methods without adaptation fail. Our method has shown 43.51%, 5.09%, and 38.62% improvement after adaptation using a deep neural network for predicting the precipitation of Paris, Los Angeles, and Tokyo, respectively.

Gaining a deeper understanding of weather and being able to predict its future conduct have always been considered important endeavors for the growth of our society. This research paper explores the advancements in understanding and predicting nature's behavior, particularly in the context of weather forecasting, through the application of machine learning algorithms. By leveraging the power of machine learning, data mining, and data analysis techniques, significant progress has been made in this field. This study focuses on analyzing the contributions of various machine learning algorithms in predicting precipitation and temperature patterns using a 20-year dataset from a single weather station in Dhaka city. Algorithms such as Gradient Boosting, AdaBoosting, Artificial Neural Network, Stacking Random Forest, Stacking Neural Network, and Stacking KNN are evaluated and compared based on their performance metrics, including Confusion matrix measurements. The findings highlight remarkable achievements and provide valuable insights into their performances and features correlation.

We introduce CroissantLLM, a 1.3B language model pretrained on a set of 3T English and French tokens, to bring to the research and industrial community a high-performance, fully open-sourced bilingual model that runs swiftly on consumer-grade local hardware. To that end, we pioneer the approach of training an intrinsically bilingual model with a 1:1 English-to-French pretraining data ratio, a custom tokenizer, and bilingual finetuning datasets. We release the training dataset, notably containing a French split with manually curated, high-quality, and varied data sources. To assess performance outside of English, we craft a novel benchmark, FrenchBench, consisting of an array of classification and generation tasks, covering various orthogonal aspects of model performance in the French Language. Additionally, rooted in transparency and to foster further Large Language Model research, we release codebases, and dozens of checkpoints across various model sizes, training data distributions, and training steps, as well as fine-tuned Chat models, and strong translation models. We evaluate our model through the FMTI framework, and validate 81 % of the transparency criteria, far beyond the scores of even most open initiatives. This work enriches the NLP landscape, breaking away from previous English-centric work in order to strengthen our understanding of multilinguality in language models.

Global climate change is a huge challenge facing society today. The rapid growth of computing overall and of machine learning (ML) in particular rightfully raises concerns about their carbon footprints. As an early and enthusiastic adopter of ML, a manufacturer of millions of smartphones annually, and a significant cloud provider, Google is in a nearly unique position to compare the impact and efficiency of ML on the two ends of the information technology (IT) computing spectrum. Keep in mind this article is not a comparison of all computation done on phones and the cloud, but solely on the impact of ML on energy use and operational CO2e. We provide the data to support these insights. While primarily focused on operational CO2e generated from computer use, we also address the relative impact of embodied CO2e. Computers in datacenters draw electricity from the grid continuously. Because smartphones operate from a battery, they only draw electricity from the grid when connected to a charger. To account for smartphone ML energy accurately, we must include the energy overhead of their chargers. Wireless charging is increasingly popular due to its convenience and the reduction in smartphone wear and tear by avoiding the repeated insertion of a cable. For wired charging, energy is lost from the AC/DC power adapter in the charger and in the power management integrated circuit (PMIC) battery charger in the phone. Wireless charging loses additional energy through the inductive coils.

Network-on-Chip (NoC) is widely used to facilitate communication between components in sophisticated System-on-Chip (SoC) designs. Security of the on-chip communication is crucial because exploiting any vulnerability in shared NoC would be a goldmine for an attacker that puts the entire computing infrastructure at risk. NoC security relies on effective countermeasures against diverse attacks, including attacks on anonymity. We investigate the security strength of existing anonymous routing protocols in NoC architectures. Specifically, this paper makes two important contributions. We show that the existing anonymous routing is vulnerable to machine learning (ML) based flow correlation attacks on NoCs. We propose lightweight anonymous routing with traffic obfuscation techniques to defend against ML-based flow correlation attacks. Experimental studies using both real and synthetic traffic reveal that our proposed attack is successful against state-of-the-art anonymous routing in NoC architectures with high accuracy (up to 99%) for diverse traffic patterns, while our lightweight countermeasure can defend against ML-based attacks with minor hardware and performance overhead.

We propose a physics-constrained convolutional neural network (PC-CNN) to solve two types of inverse problems in partial differential equations (PDEs), which are nonlinear and vary both in space and time. In the first inverse problem, we are given data that is offset by spatially varying systematic error (i.e., the bias, also known as the epistemic uncertainty). The task is to uncover from the biased data the true state, which is the solution of the PDE. In the second inverse problem, we are given sparse information on the solution of a PDE. The task is to reconstruct the solution in space with high-resolution. First, we present the PC-CNN, which constrains the PDE with a simple time-windowing scheme to handle sequential data. Second, we analyse the performance of the PC-CNN for uncovering solutions from biased data. We analyse both linear and nonlinear convection-diffusion equations, and the Navier-Stokes equations, which govern the spatiotemporally chaotic dynamics of turbulent flows. We find that the PC-CNN correctly recovers the true solution for a variety of biases, which are parameterised as non-convex functions. Third, we analyse the performance of the PC-CNN for reconstructing solutions from biased data for the turbulent flow. We reconstruct the spatiotemporal chaotic solution on a high-resolution grid from only 2\% of the information contained in it. For both tasks, we further analyse the Navier-Stokes solutions. We find that the inferred solutions have a physical spectral energy content, whereas traditional methods, such as interpolation, do not. This work opens opportunities for solving inverse problems with partial differential equations.

In particular, PCAC is implemented as a cold-start A fundamental necessity for autonomous atmospheric indirect adaptive controller, where the plant model order is flight vehicles is a reliable autopilot for controlling the specified as a hyperparameter, but otherwise no plant model attitude and flight path. For a fixed-wing vehicle, stability and is assumed to be available. The identified model updated control derivatives are typically determined through windtunnel by RLS is linear, and thus it is suitable for modeling the testing or computational modeling over a range of aircraft dynamics near trim. In practice, an autopilot designed Mach number, angle of attack, and sideslip angle. This to operate over a wide range of flight conditions depends modeling data is then used to develop an autopilot based on on gain scheduling of multiple linear controllers. The goal gain scheduling, feedback linearization, or dynamic inversion of this study is to investigate, via numerical experiments, [1], [2]. In practice, however, the aerodynamics of an aircraft the viability and potential performance of PCAC under may be too expensive to model with high accuracy or may conditions of high uncertainty, in effect, no prior modeling change due to atmospheric conditions, such as icing, as well information, without the need for gain scheduling.

In recent years, deep learning has gained increasing popularity in the fields of Partial Differential Equations (PDEs) and Reduced Order Modeling (ROM), providing domain practitioners with new powerful data-driven techniques such as Physics-Informed Neural Networks (PINNs), Neural Operators, Deep Operator Networks (DeepONets) and Deep-Learning based ROMs (DL-ROMs). In this context, deep autoencoders based on Convolutional Neural Networks (CNNs) have proven extremely effective, outperforming established techniques, such as the reduced basis method, when dealing with complex nonlinear problems. However, despite the empirical success of CNN-based autoencoders, there are only a few theoretical results supporting these architectures, usually stated in the form of universal approximation theorems. In particular, although the existing literature provides users with guidelines for designing convolutional autoencoders, the subsequent challenge of learning the latent features has been barely investigated. Furthermore, many practical questions remain unanswered, e.g., the number of snapshots needed for convergence or the neural network training strategy. In this work, using recent techniques from sparse high-dimensional function approximation, we fill some of these gaps by providing a new practical existence theorem for CNN-based autoencoders when the parameter-to-solution map is holomorphic. This regularity assumption arises in many relevant classes of parametric PDEs, such as the parametric diffusion equation, for which we discuss an explicit application of our general theory.