Added placeholder for Robust AI

Co-Authored-By: Zishen Wan <[email protected]>

_quarto.yml

@@ -72,6 +72,7 @@ book:
     - generative_ai.qmd
     - ai_for_good.qmd
     - sustainable_ai.qmd
+    - robust_ai.qmd
 
     - part: EXERCISES
       chapters: 

robust_ai.qmd

@@ -1,3 +1,50 @@
 # Robust AI
 
-Explanation: Focus on how/why robustness of the ML system is important and to what degree there is freedom to leverage approximation resilience in the system.
+Robust AI focuses on ensuring that AI systems operate reliably and safely in real-world environments. As AI systems are deployed in critical applications, from healthcare to autonomous vehicles, it's essential that they can handle unexpected situations, adversarial attacks, and hardware or software faults. This chapter delves into the various challenges and solutions associated with building robust AI systems, especially for TinyML.
+
+::: {.callout-tip collapse="true"}
+## Learning Objectives
+
+* coming soon.
+
+:::
+
+# Hardware Resilience
+
+Explanation: With the proliferation of TinyML on edge devices, the hardware on which these models run can be exposed to various environmental factors and wear-and-tear. Ensuring hardware resilience is crucial to maintain consistent AI performance.
+
+## Compute Faults
+
+Description: Discusses issues related to faults in the computation units, such as CPUs, GPUs, and custom accelerators. This can include issues like overheating, transistor failures, or other malfunctions.
+
+## Memory Faults
+
+Description: Addresses faults in the memory components of a system, including RAM, cache, and storage. Topics can include bit flips, wear-out, and other memory-related issues.
+
+# Software Resilience
+
+Explanation: Software forms the backbone of any AI system. Ensuring its resilience means that the system can handle unexpected inputs, software bugs, or other issues without catastrophic failure.
+
+## Framework Faults
+
+Description: Discusses potential issues in the software stack, from the OS to the AI framework. This can include bugs, version incompatibilities, or other software-related problems.
+
+## Faulty Inputs
+
+Description: Explores how AI systems can handle unexpected or corrupted inputs. This is especially important for systems in the real world where input data can be noisy or unreliable.
+
+# Model Resilience
+
+As the core of any AI system, the model's resilience to various challenges, from adversarial attacks to real-world data shifts, is paramount for reliable operation.
+
+## Worst-case Faults
+
+Description: Investigates the model's behavior under worst-case scenarios, such as extreme data values or conditions outside the training distribution.
+
+## Adversarial Attacks 
+
+Description: Discusses potential threats where malicious actors intentionally manipulate inputs to deceive the AI model, and strategies to defend against these attacks.
+
+# Conclusion
+
+Explanation: Conclude with the key highlights.