Machine learning models, such as ChatGPT and those used in autonomous driving, are becoming essential tools in our daily lives. However, the existence of Adversarial Examples demonstrates that these systems are not free from vulnerabilities. To ensure their reliability, it is crucial to proactively address the potential risks associated with their use in critical safety applications.

In a recent blog post, relAI PhD student  Lukas Gosch introduces the concept of Adversarial Examples and discusses Certifiable Robustness, a methodology designed to combat 🛡️ them.

What is an Adversarial Example?

As Lukas Gosch outlines, Adversarial Examples are deliberately crafted inputs that cause machine learning models to misclassify data. For example, the strategic placement of stickers on traffic signs can lead to incorrect identification of road signs by machine learning systems used in autonomous vehicles. Additionally, if an adversary manipulates the training data upon which these models are built, this too qualifies as an Adversarial Example.

How to combat Adversarial Examples?

In his post, Lukas describes Certifiable Robustness, a methodology for verifying the resilience of machine learning systems against adversarial examples, and explores the challenges associated with it.

👉 Check it out https://zuseschoolrelai.de/blog/a-beginners-guide-to-certifiable-robustness/

Ethics training in the field of Artificial Intelligence (AI) is essential. AI is revolutionizing our world and transforming various aspects of society, including education, politics, and economics. Researchers in AI are facilitating these changes by developing the conceptual and technical foundation 🧱 for new social structures.

🤔 However, are AI researchers fully aware of their responsibility and the impact of their work on future society?

To encourage reflection on these issues and provide foundational ethics training, relAI recently introduced an Ethics course as a mandatory part of the curriculum. The course is taught by Prof. Ruth Müller from the Department of Science, Technology, and Society at the TUM School of Social Sciences and Technology at Technische Universität München. The first edition took place last winter semester and was well received by relAI students.

One of these students, Valentine Idakwo, has written an insightful blog post to share the lessons learned from the course. This engaging lecture encourages students of AI to analyze their work from an external perspective, considering the social context in which their work is embedded.

👉 Check it out! https://zuseschoolrelai.de/blog/social-impact-ai-research-relai/

📋Textual generative models, such as the GPT family, have created new opportunities for human–AI interaction. They possess impressive abilities to summarize lengthy documents, compose poetry, and answer complex questions. However, alongside these remarkable capabilities lies a significant challenge: 🪄 hallucinations.

🔹What are hallucinations❓

Hallucinations are instances in which a model generates content that is factually incorrect, lacks supporting evidence, or is entirely fabricated.

🔹Why is preventing hallucination important❓

The implications of hallucination can be severe in various real-world domains. In healthcare, hallucinated outputs might suggest non-existent treatments, potentially placing patients at risk. In legal contexts, fabricated precedents could mislead practitioners and affect judicial outcomes. Similarly, in journalism, factual errors in AI-generated articles could lead to misinformation and erode public trust in media institutions. These examples underscore that hallucination is not just a technical flaw; it is a serious societal concern.

In his first post, relAI PhD student Bailan He explains various scenarios that may lead to hallucinations in generative models. He discusses the approaches developed to detect and mitigate these issues.

This post provides a basic summary of strategies designed to ensure that generative models fulfill their essential responsibility: ✨ producing truthful content ✨.

👉 Do not miss it!  https://zuseschoolrelai.de/blog/responsible-textual-generative-models-part-i-generating-truthful-content/

📢 New relAI blog post!

AI models have recently revolutionized medical imaging by enabling automated analysis of complex radiological data. This includes tasks such as lesion detection, organ segmentation, and disease progression prediction. However, those models often overfit, learning patterns specific to the training dataset rather than acquiring generalizable visual concepts. In this blog post, relAI PhD student Aswathi introduces Random Convolutions, a method designed to enhance the generalization capabilities of AI models applied to medical images.

👉 https://zuseschoolrelai.de/blog/random-convolutions-a-simple-way-to-boost-generalization/

The power consumption of the brain is approximately 20 watts, comparable to that of some light bulbs 💡. Currently, digital computers operate at power inputs that are several orders of magnitude higher, particularly when AI algorithms attempt to emulate a subset of brain capabilities.

🪄Should we consider mimicking the brain to investigate the potential in biomimetic computation and algorithms, especially in the design of intelligent and efficient robotic agents?  

In this blog post, relAI PhD student Ahmed Abdelrahman, introduces neuromorphic computing. This emerging research field aims to replicate the brain's fundamental neural structures and characteristics in silico, paving the way for the next generation of AI.

It is definitely a must-read post: https://zuseschoolrelai.de/blog/neuromorphic-computing/.