The paper introduces LLMEffiChecker, a tool to test the computational efficiency robustness of LLMs by identifying vulnerabilities that can significantly degrade performance. LLMEffiChecker uses both white-box (gradient-guided perturbation) and black-box (causal inference-based perturbation) methods to delay the generation of the end-of-sequence token. Experiments on nine public LLMs demonstrate that LLMEffiChecker can substantially increase response latency and energy consumption with minimal input perturbations.
The paper proposes a method for causal inference using satellite image time series to determine the impact of interventions on climate change, focusing on quantifying deforestation due to human causes. The method uses computer vision and deep learning to detect forest tree coverage levels over time and Bayesian structural causal models to estimate counterfactuals. The framework is applied to analyze deforestation levels before and after the hyperinflation event in Brazil in the Amazon rainforest region.
MBZUAI and UC Berkeley held a joint workshop on machine learning, featuring discussions on online learning, fair allocation in dynamic mechanism design, and causal inference. Michael I. Jordan, Laureate Professor and Honorary Program Director at MBZUAI, highlighted the institute's rapid growth during his visit. Researchers explored methods for enhancing the properties of large, complex models, such as calibration, fairness, and robustness. Why it matters: Such collaborations advance AI research and foster knowledge exchange between leading global experts and regional institutions like MBZUAI.
This article discusses the use of artificial intelligence in precision oncology, particularly in understanding individual tumor mechanisms and aiding clinical decision-making. Dr. Xinghua Lu, with extensive experience in medicine and biomedical informatics, will present research on individualized Bayesian causal inference methods for investigating oncogenic mechanisms. These methods aim to provide clinical decision support at the cellular, tumor, and patient levels. Why it matters: AI-driven precision oncology can enable more personalized and effective cancer treatments, improving patient outcomes in the region and globally.
Researchers are exploring methods for evaluating the outcome of actions using off-policy observations where the context is noisy or anonymized. They employ proxy causal learning, using two noisy views of the context to recover the average causal effect of an action without explicitly modeling the hidden context. The implementation uses learned neural net representations for both action and context, and demonstrates outperformance compared to an autoencoder-based alternative. Why it matters: This research addresses a key challenge in applying AI in real-world scenarios where data privacy or bandwidth limitations necessitate working with noisy or anonymized data.
MBZUAI hosted a talk on causal AI, featuring Professor Jin Tian from Iowa State University. The talk covered enriching AI systems with causal reasoning capabilities, moving AI beyond prediction to understanding. Professor Tian shared research on causal inference and estimating causal effects from data, using a novel estimator with double/debiased machine learning (DML) properties. Why it matters: Causal AI can improve the explainability, robustness, and adaptability of AI systems, addressing limitations of purely statistical models.
MBZUAI researchers presented a new causal discovery method at NeurIPS that identifies relationships between deterministic and non-deterministic variables. The method builds directed graphs visualizing relationships between variables, incorporating both probabilistic and deterministic principles. The lead author, Longkang Li, aims to apply causal discovery to healthcare and biology for better understanding of diseases. Why it matters: This research advances the field of causal inference, potentially improving applications in areas like healthcare where understanding complex relationships is critical.
MBZUAI researchers presented a study at ICML 2024 examining how data aggregation distorts causal discovery. The study argues that current methods are misled because real-world interactions happen at a micro level while observations are aggregated. Using the example of ice cream sales and temperature, they highlight how aggregation introduces "instantaneous causality" where time-lags exist. Why it matters: The research identifies a fundamental limitation in current causal discovery methods, potentially impacting disciplines relying on accurate causal inference from observational data.