Deep neural networks trained with different architectures, objectives, and datasets have been reported to converge on similar visual representations. However, what remains unknown is which visual properties models actually converge on and which factors may underlie this convergence. To address this, we decompose the object similarity structure of 162 diverse vision models into a small set of non-negative dimensions. To determine universal versus model-specific dimensions, we then estimate how often each dimension reappears across models. In contrast to model-specific dimensions, universal dimensions are more interpretable and more strongly driven by conceptual image properties, indicating the relevance of interpretability and semantic content as implicit factors driving universality across models. Differences in architecture, objective function, training data, model size, and model performance do not explain the emergence of universal dimensions. However, models with more universal dimensions also better predict macaque IT activity and human similarity judgments, suggesting that universality reflects representations relevant to biological vision. These findings have important implications for understanding the emergent representations underlying deep neural network models and their alignment with biological vision.
Francisco Pereira, Justine Sherry, Pratiksha Thaker
原文摘要
In 2003, Clark et al. proposed the "Knowledge Plane", a layer of intelligent agents that share rich information about network state in order to collaboratively manage the Internet across administrative boundaries. The technology of the day lacked the two capabilities it demanded most: reasoning and unstructured communication. Thus, automated network management has remained largely confined to single domains, while interdomain problems are still resolved by human operators on the phone. We argue that agentic large language models can finally unlock the potential of the Knowledge Plane, but that doing so will require solving new research problems that arise particularly in this decentralized, multi-agent, coopetitive setting that have received limited attention in the literature thus far.
João Romeiras Amado, Francisco Pereira, David Pissarra, Salvatore Signorello, Miguel Correia, Fernando M. V. Ramos
原文摘要
Machine Learning (ML)-based malicious traffic detectors can identify unseen attacks, including in encrypted traffic, but scaling them to Terabit rates remains challenging. Packet sampling, commonly used to meet processing limits, can discard essential behavioral information and render detection ineffective. We present Peregrine, a hardware-software co-designed architecture built around a simple principle: compute features before sampling. Peregrine updates ML features over all traffic at line rate in a programmable switch, then downsamples the feature records sent to software inference. We instantiate this design with two detectors that impose distinct feature-computation requirements on the switch. Across 84 attack traces from three datasets, we show that approximate features computed over all packets can outperform exact features computed from sampled packets. In our evaluation, Peregrine preserves detection effectiveness while reducing feature-record rates by three orders of magnitude - even where pre-feature sampling causes detection to collapse.