We carry out a series of experiments to test large language models' multi-hop reasoning ability from three aspects: selecting and combining external knowledge, dealing with non-sequential reasoning tasks and generalising to data samples with larger numbers of hops. We test the GPT-3.5 model on four reasoning benchmarks with Chain-of-Thought prompting (and its variations). Our results reveal that despite the amazing performance achieved by large language models on various reasoning tasks, models still suffer from severe drawbacks which shows a large gap with humans.

Wind power forecasting plays a critical role in modern energy systems, facilitating the integration of renewable energy sources into the power grid. Accurate prediction of wind energy output is essential for managing the inherent intermittency of wind power, optimizing energy dispatch, and ensuring grid stability. This paper proposes the use of Deep Neural Network (DNN)-based predictive models that leverage climate datasets, including wind speed, atmospheric pressure, temperature, and other meteorological variables, to improve the accuracy of wind power simulations. In particular, we focus on the Coupled Model Intercomparison Project (CMIP) datasets, which provide climate projections, as inputs for training the DNN models. These models aim to capture the complex nonlinear relationships between the CMIP-based climate data and actual wind power generation at wind farms located in Germany. Our study compares various DNN architectures, specifically Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) networks, and Transformer-enhanced LSTM models, to identify the best configuration among these architectures for climate-aware wind power simulation. The implementation of this framework involves the development of a Python package (CADNN) designed to support multiple tasks, including statistical analysis of the climate data, data visualization, preprocessing, DNN training, and performance evaluation. We demonstrate that the DNN models, when integrated with climate data, significantly enhance forecasting accuracy. This climate-aware approach offers a deeper understanding of the time-dependent climate patterns that influence wind power generation, providing more accurate predictions and making it adaptable to other geographical regions.

Electric Power Steering (EPS) systems utilize electric motors to aid users in steering their vehicles, which provide additional precise control and reduced energy consumption compared to traditional hydraulic systems. EPS technology provides safety,control and efficiency.. This paper explains the integration of Artificial Intelligence (AI) into Electric Power Steering (EPS) systems, focusing on its role in enhancing the safety, and adaptability across diverse driving conditions. We explore significant development in AI-driven EPS, including predictive control algorithms, adaptive torque management systems, and data-driven diagnostics. The paper presents case studies of AI applications in EPS, such as Lane centering control (LCC), Automated Parking Systems, and Autonomous Vehicle Steering, while considering the challenges, limitations, and future prospects of this technology. This article discusses current developments in AI-driven EPS, emphasizing on the benefits of improved safety, adaptive control, and predictive maintenance. Challenges in integrating AI in EPS systems. This paper addresses cybersecurity risks, ethical concerns, and technical limitations,, along with next steps for research and implementation in autonomous, and connected vehicles.

For end-to-end autonomous driving (E2E-AD), the evaluation system remains an open problem. Existing closed-loop evaluation protocols usually rely on simulators like CARLA being less realistic; while NAVSIM using real-world vision data, yet is limited to fixed planning trajectories in short horizon and assumes other agents are not reactive. We introduce Bench2Drive-R, a generative framework that enables reactive closed-loop evaluation. Unlike existing video generative models for AD, the proposed designs are tailored for interactive simulation, where sensor rendering and behavior rollout are decoupled by applying a separate behavioral controller to simulate the reactions of surrounding agents. As a result, the renderer could focus on image fidelity, control adherence, and spatial-temporal coherence. For temporal consistency, due to the step-wise interaction nature of simulation, we design a noise modulating temporal encoder with Gaussian blurring to encourage long-horizon autoregressive rollout of image sequences without deteriorating distribution shifts. For spatial consistency, a retrieval mechanism, which takes the spatially nearest images as references, is introduced to to ensure scene-level rendering fidelity during the generation process. The spatial relations between target and reference are explicitly modeled with 3D relative position encodings and the potential over-reliance of reference images is mitigated with hierarchical sampling and classifier-free guidance. We compare the generation quality of Bench2Drive-R with existing generative models and achieve state-of-the-art performance. We further integrate Bench2Drive-R into nuPlan and evaluate the generative qualities with closed-loop simulation results. We will open source our code.

Large Language Models (LLMs) are trained on extensive datasets that often contain sensitive, human-generated information, raising significant concerns about privacy breaches. While certified unlearning approaches offer strong privacy guarantees, they rely on restrictive model assumptions that are not applicable to LLMs. As a result, various unlearning heuristics have been proposed, with the associated privacy risks assessed only empirically. The standard evaluation pipelines typically randomly select data for removal from the training set, apply unlearning techniques, and use membership inference attacks (MIAs) to compare the unlearned models against models retrained without the to-be-unlearned data. However, since every data point is subject to the right to be forgotten, unlearning should be considered in the worst-case scenario from the privacy perspective. Prior work shows that data outliers may exhibit higher memorization effects. Intuitively, they are harder to be unlearn and thus the privacy risk of unlearning them is overlooked and underestimated in the current evaluation. In this paper, we leverage minority data to identify such a critical flaw in previously widely adopted evaluations. We substantiate this claim through carefully designed experiments, including unlearning canaries related to minority groups, inspired by privacy auditing literature. Using personally identifiable information (PII) as a representative minority identifier, we demonstrate that minority groups experience at least 20% more privacy leakage in most cases across six unlearning approaches, three MIAs, three benchmark datasets, and two LLMs of different scales. Given that the right to be forgotten should be upheld for every individual, we advocate for a more rigorous evaluation of LLM unlearning methods. Our minority-aware evaluation framework represents an initial step toward ensuring more equitable and thorough assessments of LLM unlearning efficacy.

Artificial Intelligence is widely regarded as a transformative force with the potential to redefine numerous sectors of human civilization. While Artificial Intelligence has evolved from speculative fiction to a pivotal element of technological progress, its role as a truly transformative agent, or transformative Artificial Intelligence, remains a subject of debate. This work explores the historical precedents of technological breakthroughs, examining whether Artificial Intelligence can achieve a comparable impact, and it delves into various ethical frameworks that shape the perception and development of Artificial Intelligence. Additionally, it considers the societal, technical, and regulatory challenges that must be addressed for Artificial Intelligence to become a catalyst for global change. We also examine not only the strategies and methodologies that could lead to transformative Artificial Intelligence but also the barriers that could ultimately make these goals unattainable. We end with a critical inquiry into whether reaching a transformative Artificial Intelligence might compel humanity to adopt an entirely new ethical approach, tailored to the complexities of advanced Artificial Intelligence. By addressing the ethical, social, and scientific dimensions of Artificial Intelligence's development, this work contributes to the broader discourse on the long-term implications of Artificial Intelligence and its capacity to drive civilization toward a new era of progress or, conversely, exacerbate existing inequalities and risks.

Currently, data-intensive scientific applications require vast amounts of compute resources to deliver world-leading science. The climate emergency has made it clear that unlimited use of resources (e.g., energy) for scientific discovery is no longer acceptable. Future computing hardware promises to be much more energy efficient, but without better optimized software this cannot reach its full potential. In this vision paper, we propose a generic AI-driven co-design methodology, using specialized Large Language Models (like ChatGPT), to effectively generate efficient code for emerging computing hardware. We describe how we will validate our methodology with two radio astronomy applications, with sustainability as the key performance indicator. This paper is a modified version of our accepted SuperCode project proposal. We present it here in this form to introduce the vision behind this project and to disseminate the work in the spirit of Open Science and transparency. An additional aim is to collect feedback, invite potential collaboration partners and use-cases to join the project.

A fundamental step in the development of machine learning models commonly involves the tuning of hyperparameters, often leading to multiple model training runs to work out the best-performing configuration. As machine learning tasks and models grow in complexity, there is an escalating need for solutions that not only improve performance but also address sustainability concerns. Existing strategies predominantly focus on maximizing the performance of the model without considering energy efficiency. To bridge this gap, in this paper, we introduce Spend More to Save More (SM2), an energy-aware hyperparameter optimization implementation based on the widely adopted successive halving algorithm. Unlike conventional approaches including energy-intensive testing of individual hyperparameter configurations, SM2 employs exploratory pretraining to identify inefficient configurations with minimal energy expenditure. Incorporating hardware characteristics and real-time energy consumption tracking, SM2 identifies an optimal configuration that not only maximizes the performance of the model but also enables energy-efficient training. Experimental validations across various datasets, models, and hardware setups confirm the efficacy of SM2 to prevent the waste of energy during the training of hyperparameter configurations.

Partial differential equations (PDEs) are fundamental mathematical frameworks used to describe complex physical phenomena across diverse scientific and engineering domains. From fluid dynamics and climate modeling to quantum mechanics and biological systems, PDEs encapsulate intricate interactions and dynamical behaviors derived from underlying physical principles. However, solving PDEs, particularly nonlinear equations with complex boundary conditions, poses significant computational challenges, historically limiting our ability to simulate and predict such systems accurately. The computational landscape for solving PDEs has been transformed by the integration of machine learning (ML) and artificial intelligence (AI) techniques. Recent advancements have introduced a proliferation of operator learning methods, each contributing unique insights and capabilities for tackling complex mathematical problems.

This paper proposes a novel loss function, called 'Tube Loss', for simultaneous estimation of bounds of a Prediction Interval (PI) in the regression setup, and also for generating probabilistic forecasts from time series data solving a single optimization problem. The PIs obtained by minimizing the empirical risk based on the Tube Loss are shown to be of better quality than the PIs obtained by the existing methods in the following sense. First, it yields intervals that attain the prespecified confidence level $t \in(0,1)$ asymptotically. A theoretical proof of this fact is given. Secondly, the user is allowed to move the interval up or down by controlling the value of a parameter. This helps the user to choose a PI capturing denser regions of the probability distribution of the response variable inside the interval, and thus, sharpening its width. This is shown to be especially useful when the conditional distribution of the response variable is skewed. Further, the Tube Loss based PI estimation method can trade-off between the coverage and the average width by solving a single optimization problem. It enables further reduction of the average width of PI through re-calibration. Also, unlike a few existing PI estimation methods the gradient descent (GD) method can be used for minimization of empirical risk. Finally, through extensive experimentation, we have shown the efficacy of the Tube Loss based PI estimation in kernel machines, neural networks and deep networks and also for probabilistic forecasting tasks. The codes of the experiments are available at https://github.com/ltpritamanand/Tube_loss