https://www.youtube.com/watch?v=IkrCax_Upa8
"inside a neural network"

Assessing Generalization of SGD via Disagreement

Abstract:

We empirically show that the test error of deep networks can be estimated by simply training the same architecture on the same training set but with a different run of Stochastic Gradient Descent (SGD), and measuring the disagreement rate between the two networks on unlabeled test data. This builds on -- and is a stronger version of -- the observation in Nakkiran & Bansal '20, which requires the second run to be on an altogether fresh training set. We further theoretically show that this peculiar phenomenon arises from the \emph{well-calibrated} nature of \emph{ensembles} of SGD-trained models. This finding not only provides a simple empirical measure to directly predict the test error using unlabeled test data, but also establishes a new conceptual connection between generalization and calibration.

https://arxiv.org/abs/2106.13799
#abstract

Walrasian equilibrium behavior in nature

Abstract:

The interaction between land plants and mycorrhizal fungi (MF) forms perhaps the world’s most prevalent biological market. Most plants participate in such markets, in which MF collect nutrients from the soil and trade them with host plants in exchange for carbon. In a recent study, M. D. Whiteside et al. [Curr. Biol. 29, 2043–2050.e8 (2019)] conducted experiments that allowed them to quantify the behavior of arbuscular MF when trading phosphorus with their host roots. Their experimental techniques enabled the researchers to infer the quantities traded under multiple scenarios involving different amounts of phosphorus resources initially held by different MF patches. We use these observations to confirm a revealed preference hypothesis, which characterizes behavior in Walrasian equilibrium, a centerpiece of general economic equilibrium theory.

https://www.pnas.org/content/118/27/e2020961118

tl;dr - fungi embrace fundamental economic theory as they engage in trading
#abstract

A topological solution to object segmentation and tracking

Abstract:

The world is composed of objects, the ground, and the sky. Visual perception of objects requires solving two fundamental challenges: segmenting visual input into discrete units, and tracking identities of these units despite appearance changes due to object deformation, changing perspective, and dynamic occlusion. Current computer vision approaches to segmentation and tracking that approach human performance all require learning, raising the question: can objects be segmented and tracked without learning? Here, we show that the mathematical structure of light rays reflected from environment surfaces yields a natural representation of persistent surfaces, and this surface representation provides a solution to both the segmentation and tracking problems. We describe how to generate this surface representation from continuous visual input, and demonstrate that our approach can segment and invariantly track objects in cluttered synthetic video despite severe appearance changes, without requiring learning.

https://arxiv.org/abs/2107.02036

In 1984 a researcher named Bloom found that students learning mastery-based and with one-on-one mentorship perform two standard deviations better than those in a conventional classroom.

Incredible to know, but too expensive to do anything about, so nothing changed.

.

>Once in the brain, the nanosensors are highly sensitive to local changes in the electric field. In laboratory tests, in vitro prototypes of the NeuroSWARM3 were able to generate a signal-to-noise ratio of over 1,000, a sensitivity level that is suitable for detecting the electrical signal generated when a single neuron fires.

https://phys.org/news/2021-07-tiny-sensors-brain-surgery-implants.html

https://slimemoldtimemold.com/2021/07/07/a-chemical-hunger-part-i-mysteries/

A bit offtop, but really good stuff

https://www.youtube.com/watch?v=60z_hpEAtD8
Old, but gold

tl;dr:
OTRv4 > OpenPGP > OMEMO

https://github.com/pepe-i-shim/thinkpad-from-ebay
more from @scp_shitpost