Predicting the Performance of Foundation Models via Agreement-on-the-Line

Estimating the out-of-distribution performance in regimes where labels are scarce is critical to safely deploy foundation models. Recently, it was shown that ensembles of neural networks observe the phenomena ``agreement-on-the-line'', which can be leveraged to reliably predict OOD performance without labels. However, in contrast to classical neural networks that are trained on in-distribution data from scratch for numerous epochs, foundation models undergo minimal finetuning from heavily pretrained weights, which may reduce the ensemble diversity needed to observe agreement-on-the-line. In our work, we demonstrate that when lightly finetuning multiple runs from a $\textit{single}$ foundation model, the choice of randomness during training (linear head initialization, data ordering, and data subsetting) can lead to drastically different levels of agreement-on-the-line in the resulting ensemble. Surprisingly, only random head initialization is able to reliably induce agreement-on-the-line in finetuned foundation models across vision and language benchmarks. Second, we demonstrate that ensembles of $\textit{multiple}$ foundation models pretrained on different datasets but finetuned on the same task can also show agreement-on-the-line. In total, by careful construction of a diverse ensemble, we can utilize agreement-on-the-line-based methods to predict the OOD performance of foundation models with high precision.

Heuristic Algorithm-based Action Masking Reinforcement Learning (HAAM-RL) with Ensemble Inference Method

This paper presents a novel reinforcement learning (RL) approach called HAAM-RL (Heuristic Algorithm-based Action Masking Reinforcement Learning) for optimizing the color batching re-sequencing problem in automobile painting processes. The existing heuristic algorithms have limitations in adequately reflecting real-world constraints and accurately predicting logistics performance. Our methodology incorporates several key techniques including a tailored Markov Decision Process (MDP) formulation, reward setting including Potential-Based Reward Shaping, action masking using heuristic algorithms (HAAM-RL), and an ensemble inference method that combines multiple RL models. The RL agent is trained and evaluated using FlexSim, a commercial 3D simulation software, integrated with our RL MLOps platform BakingSoDA. Experimental results across 30 scenarios demonstrate that HAAM-RL with an ensemble inference method achieves a 16.25% performance improvement over the conventional heuristic algorithm, with stable and consistent results. The proposed approach exhibits superior performance and generalization capability, indicating its effectiveness in optimizing complex manufacturing processes. The study also discusses future research directions, including alternative state representations, incorporating model-based RL methods, and integrating additional real-world constraints.

Jailbreaking is Best Solved by Definition

The rise of "jailbreak" attacks on language models has led to a flurry of defenses aimed at preventing the output of undesirable responses. In this work, we critically examine the two stages of the defense pipeline: (i) the definition of what constitutes unsafe outputs, and (ii) the enforcement of the definition via methods such as input processing or fine-tuning. We cast severe doubt on the efficacy of existing enforcement mechanisms by showing that they fail to defend even for a simple definition of unsafe outputs--outputs that contain the word "purple". In contrast, post-processing outputs is perfectly robust for such a definition. Drawing on our results, we present our position that the real challenge in defending jailbreaks lies in obtaining a good definition of unsafe responses: without a good definition, no enforcement strategy can succeed, but with a good definition, output processing already serves as a robust baseline albeit with inference-time overheads.

Toward Reinforcement Learning-based Rectilinear Macro Placement Under Human Constraints

Macro placement is a critical phase in chip design, which becomes more intricate when involving general rectilinear macros and layout areas. Furthermore, macro placement that incorporates human-like constraints, such as design hierarchy and peripheral bias, has the potential to significantly reduce the amount of additional manual labor required from designers. This study proposes a methodology that leverages an approach suggested by Google's Circuit Training (G-CT) to provide a learning-based macro placer that not only supports placing rectilinear cases, but also adheres to crucial human-like design principles. Our experimental results demonstrate the effectiveness of our framework in achieving power-performance-area (PPA) metrics and in obtaining placements of high quality, comparable to those produced with human intervention. Additionally, our methodology shows potential as a generalized model to address diverse macro shapes and layout areas.