Robustness and scaling

Does Robustness Improve with Scale? | FAR.AI

We find that scale alone does little to improve model robustness, but that larger models benefit more from defenses such as adversarial training than do ...

Robustness to Scale — LessWrong

Robustness to relative scale means that your AI system does not depend on any subsystems being similarly powerful to each other. This is most ...

Scaling Compute Is Not All You Need for Adversarial Robustness

Abstract page for arXiv paper 2312.13131: Scaling Compute Is Not All You Need for Adversarial Robustness.

Robustness to Scale — AI Alignment Forum

Robustness to scaling up means that your AI system does not depend on not being too powerful. One way to check for this is to think about what ...

[2407.18213] Effects of Scale on Language Model Robustness - arXiv

Title:Effects of Scale on Language Model Robustness ... Abstract:Language models exhibit scaling laws, whereby increasing model and dataset size ...

Robust measures of scale - Wikipedia

Robust measures of scale ... In statistics, robust measures of scale are methods that quantify the statistical dispersion in a sample of numerical data while ...

Robust Scaling: Why and How to Use It to Handle Outliers

This post will introduce robust scaling that works well on features with outliers. Then we'll discuss why standard scaling succumbs to outliers.

Feature Scaling Using Standardization, Normalization and Robust ...

Robust Scaling. This method uses the interquartile range (Q3 — Q1) to adjust the range of features. It works well if you have a lot of outliers ...

Robustness to Scaling Down: More Important Than I Thought

Robustness to scaling down, meaning that a solution to alignment keeps working if the AI is not optimal or perfect. Robustness to relative scale ...

Exploring Scaling Trends in LLM Robustness - FAR.AI

Prior work indicates that computer vision models become more robust with model and data scaling, raising the question: does language model ...

Revisiting Group Robustness: Class-specific Scaling is All You Need

We first propose a simple class-specific scaling strategy to control the trade-off between robust and average accuracies flexibly and efficiently.

Robustness scaling in large networks - IEEE Xplore

Robustness scaling in large networks ... Abstract: We propose new asymptotic notions that capture the robustness of a large interconnected network as dimension of ...

Adversarial Robustness Limits via Scaling-Law and Human ...

Our scaling laws reveal inefficiencies in prior art and provide actionable feedback to advance the field. For instance, we discovered that SOTA ...

Database scaling techniques for robustness and resilience - Medium

Database scaling techniques for robustness and resilience · Add more database nodes: This refers to adding more database servers to support ...

Optimality, fairness, and robustness in speed scaling designs

It is possible to be near-optimal and robust using SRPT scheduling and dynamic speed scaling, but this creates unfairness. SRPT can be fair and still near-.

Test scores' robustness to scaling: The scale_transformation command

The command assesses the robustness of an achievement gap between two subgroups to any arbitrary choice of scale by finding bounds for the original gap ...

Scaling LLMs For Robustness | Restackio

Scaling LLMs for Enhanced Robustness Against Adversarial Attacks ... Language models have shown that scaling their size and training data can lead ...

Studying the robustness of scaling methods

Studying the robustness of scaling methods. Published online by Cambridge University Press: 01 July 2016. Robin Sibson.

Exploring Scaling Trends in LLM Robustness - NASA/ADS

Exploring Scaling Trends in LLM Robustness ... Abstract. Language model capabilities predictably improve from scaling a model's size and training data. Motivated ...

Test scores' robustness to scaling: The scale_transformation command

Additionally, it finds scale transformations that are very likely and unlikely to benchmark against the results obtained. Finally, it also allows the user to ...