The team created a virtual node graph neural network (VGNN) by adding a series of flexible virtual nodes to the fixed crystal structure to represent phonons. The virtual nodes enable the output of the neural network to vary in size, so it isn’t restricted by the fixed crystal
@mit_csail
-
VGNN Method Improves Phonon Dispersion Relation Calculations Efficiently
By
–
Since it has virtual nodes to represent phonons, the VGNN can skip many complex calculations when estimating phonon dispersion relations, which makes the method more efficient than a standard GNN.
-
MIT’s ML Framework Predicts Phonon Dispersion 1000x Faster
By
–
Roughly 70% of the energy generated worldwide ends up as waste heat, according to some estimates — with the trouble coming from phonons. MIT’s new ML framework can predict phonon dispersion relations up to 1,000x faster than other AI-based techniques w/comparable or even better
-
Heat Management Innovation Boosts Energy and Electronics Efficiency
By
–
This method could help engineers design energy generation systems that produce more power, more efficiently. It could also help develop more efficient microelectronics, since managing heat is still a bottleneck to speeding up electronics.
-
Python consumes 76x more energy than C language
By
–
The energy consumption of different programming languages (v/
@burkov
): https://
bit.ly/3WiXC9M One finding: Python consumes 76x more energy while being 72x slower than C. -
Controlling Complexity: The Essence of Computer Programming
By
–
"Controlling complexity is the essence of computer programming." — Brian Kernighan
-
MIT releases MAIA multimodal interpretability analysis tool
By
–
Web: …
https://mult
imodal-interpretability.csail.mit.edu/maia/
Experiment browser: …
https://mult
imodal-interpretability.csail.mit.edu/maia/experimen
t-browser/
…
Paper: https://
shorturl.at/kx74G Lead authors: Tamar Rott Shaham (
@TamarRottShaham
) & Sarah Schwettmann (
@cogconfluence
)
Senior authors: Jacob Andreas (
@jacobandreas
) & Antonio Torralba -
MAIA Tool Reveals Gender Bias in AI Vision Classifiers
By
–
MAIA can reveal biases in AI systems. For example, when MAIA is prompted to find biases in the "beer glass" class of a pre-trained classifier, it reveals a gender bias; the classifier detects beer glasses held by men w/higher accuracy compared to those held by women. pic.twitter.com/fd4KTBF4EN
— MIT CSAIL (@MIT_CSAIL) 2 août 2024MAIA can reveal biases in AI systems. For example, when MAIA is prompted to find biases in the "beer glass" class of a pre-trained classifier, it reveals a gender bias; the classifier detects beer glasses held by men w/higher accuracy compared to those held by women.
-
CSAIL Method Outperforms Baselines in Neural Neuron Description
By
–
The CSAIL-led method outperformed baseline methods describing individual neurons in a variety of vision models & a new dataset of synthetic neurons w/known ground-truth descriptions.
-
MAIA: Interpretability Research Framework for AI Bias Detection
By
–
Given a query from the user (e.g. "Are biases present in my system?"), MAIA acts as an interpretability researcher: it generates hypotheses, designs experiments to test them, and refines its understanding through iterative analysis until it can solve the interpretability task.