Interesting article covered in the New York Times that describes experiments by Anthropic to understand what's really going on inside of neural network-based Large Language Models (LLMs). Researchers identified clusters of millions of neurons related to "concepts." Although neural networks are "inspired by the human brain" and are not an actual brain, the ability to excise these clusters to remove a behavior or to stimulate/focus on them to emphasize a concept does sound a lot like brain surgery. The ability to identify such clusters does kind of sound like Functional MRI. It will be interesting to see whether research into a computer model that generates language teaches us something about our own biology. Because of course we are totally not just computer code running on someone's computer. https://lnkd.in/eEM_fEzX
Patrick Biltgen’s Post
More Relevant Posts
-
Welcome to read and share the newly published paper "A Lightweight Graph Neural Network Algorithm for Action Recognition Based on Self-Distillation". This excellent paper is written by Miao Feng and Jean Meunier Via the link: https://lnkd.in/gDHSwQBs MDPI, Université de Montréal #humanactionrecognition #selfdistillation #GNN #modelcompression
A Lightweight Graph Neural Network Algorithm for Action Recognition Based on Self-Distillation
mdpi.com
To view or add a comment, sign in
-
#AgricultureMdpi – Editor's Choice Article 🔺 Impacts of Background Removal on Convolutional Neural Networks for Plant Disease Classification In-Situ ✍️ by Kamal KC et al. 🖱️ Click to read the full article: https://lnkd.in/gjxNjrXA #segmentation #backgroundsubtraction #transferlearning #deepconvolutionalneuralnetwork #plantdiseaseimageclassification
To view or add a comment, sign in
-
-
enabling digital services for Student Loan related activities while maintaining the highest security standard, the most compliant personal data protection and customer-centric data-driven innovation.
Exciting new research on deep neural network (DNN) ownership verification techniques! In our latest blog post, we present the groundbreaking study on DNN fingerprint removal attacks. We investigate the feasibility and potential impact of these attacks, and introduce RemovalNet, a min-max bilevel optimization-based attack designed to evade ownership verification. By categorizing the knowledge contained in a DNN model into general semantic and fingerprint-specific knowledge, RemovalNet effectively removes the latter while maintaining surrogate model performance. We conducted extensive experiments, assessing the fidelity, effectiveness, and efficiency of RemovalNet against four state-of-the-art defense methods on six metrics. The results are impressive: RemovalNet significantly increases the model distance between target and surrogate models by x100 compared to baseline attacks. Moreover, it achieves high efficiency, utilizing only 0.2% of the substitute dataset and 1,000 iterations. In terms of computational resources, it outperforms advanced model stealing attacks by saving up to 85%. Notably, the created surrogate model maintains high accuracy even after the DNN fingerprint removal process. Curious to learn more? Check out the blog post here: https://bit.ly/3QUaAcV. For those interested in exploring further, the RemovalNet code is readily accessible at https://bit.ly/3ssnNj6. Stay ahead of the game and delve into the fascinating realm of DNN ownership verification and removal attacks. #DNNs #FingerprintRemoval #OwnershipVerification
To view or add a comment, sign in
-
Graph neural networks (GNNs) are accelerating progress across an array of disciplines — from discovering new antibiotics and identifying drug-repurposing candidates to modeling physical systems and generating new molecules. So how do GNNs work and what other possibilities exist for their application? First authors Gabriele Corso and Hannes Stärk explain in a new Nature Reviews Methods Primers paper. Free download for a week: go.nature.com/4c6gxvr #aiforgood #aiforhealth #machinelearningsolutions #drugdiscovery #computationalchemistry
To view or add a comment, sign in
-
-
Common Vector Approach (CVA) is a known linear regression-based classifier, which also enables an extension to inter-class discrimination, known as the Discriminative Common Vector Approach (DCVA). The characteristics of linear regression classifiers (LRCs) enable the possibility of a schematic implementation that is similar to the neuron model of artificial neural networks (ANNs). In our latest research as the Pattern Recognition Research Lab. team, we explored this schematic similarity to develop an ANN representation for both CVA and DCVA. The new representation eliminates the need for projection matrices in its implementation, hence significantly reduces the memory requirements and computational complexities of the processes. Furthermore, since the new representation is in a neural style, it is expected to provide a solid and intriguing extension of CVA (and DCVA) by further incorporating adaptation or activation processes to the already successful CVA-based classifiers. This research is published in Digital Signal Processing journal: https://lnkd.in/eg4JUxAX #machinelearning #classification #neuralnetworks #regression #vectors
To view or add a comment, sign in
-
-
Recently published a wonderful paper called "Unveiling Rotational Structure in Neural Data", which I enjoy doing very much. If you ever heard about "rotational neural dynamics" and "emergent activity of motor cortex" you should definitely take a look at the paper😊😊😊 Link: https://lnkd.in/dyaxjXya
Unveiling Rotational Structure in Neural Data
biorxiv.org
To view or add a comment, sign in
-
📜 New Research Alert! An uncertainty estimation framework proposed by Guy Bergel, David Montes de Oca Zapiain, Ph.D., Vicente Romero that leverages ensembles of Bayesian neural networks to produce diverse sets of predictions using a hybrid FEM-NN approach that approximates internal forces on a deforming solid body. Discover how this framework offers a powerful tool for assessing the reliability of physics-based surrogate models by establishing uncertainty estimates for predictions spanning a wide range of possible load cases. Link: [https://lnkd.in/eBVcK2AV]
To view or add a comment, sign in
-
-
enabling digital services for Student Loan related activities while maintaining the highest security standard, the most compliant personal data protection and customer-centric data-driven innovation.
Excited to share a new blog post on CNN-Based Structural Damage Detection using Time-Series Sensor Data! Structural Health Monitoring (SHM) plays a crucial role in evaluating structural condition, aiming to detect damage through sensor data analysis. In this post, we introduce an innovative approach utilizing a Convolutional Neural Network (CNN) algorithm to enhance discrimination capabilities for damage detection. By combining spatial and temporal features, the new CNN algorithm achieves high accuracy in identifying structural degradation. Check out the full post at https://bit.ly/3si9Ww9 to learn more. #StructuralHealthMonitoring #CNNAlgorithm #DamageDetection
To view or add a comment, sign in
-
A practical and simple way to address an important problem in estimating brain connectivity from fMRI. Current methods are predominantly based on simple correlation measures, and are very susceptible to noise and variation in the signal across scans. A machine learning approach is an obvious (once someone does it first) way to gain stability. "Novel machine learning approaches for improving the reproducibility and reliability of functional and effective connectivity from functional MRI" in the Journal of Neural Engineering https://lnkd.in/gFVBvHHy
To view or add a comment, sign in
-
Co-Founder, Chief AI & Analytics Advisor @ InstaDataHelp | Innovator and Patent-Holder in Gen AI and LLM | Data Science Thought Leader and Blogger | FRSS(UK) FSASS FRIOASD | 16+ Years of Excellence
Mean Field Optimization Problem: An Exploration of Entropic Fictitious Play Entropic Fictitious Play for Mean Field Optimization Problem By Fan Chen, Zhenjie Ren, and Songbo Wang; Volume 24, Issue 211, Pages 1-36, 2023. Abstract This study focuses on the mean field limit of two-layer neural networks, where the number of neurons tends to infinity. In this scenario, the optimization of neuron parameters becomes an optimization problem over probability measures. By introducing an entropic regularizer, we identify the minimizer of the problem as a fixed point. To recover this fixed point, we propose a novel training algorithm called entropic fictitious play. This algorithm is inspired by the classical fictitious play in game theory, which is used to learn Nash equilibriums. The entropic fictitious play algorithm exhibits a two-loop iteration structure. In this paper, we prove exponential convergence and validate our theoretical results through simple numerical examples. https://lnkd.in/d5i6sphV
Mean Field Optimization Problem: An Exploration of Entropic Fictitious Play
https://instadatahelpainews.com
To view or add a comment, sign in
Thought leader in AI/ML, DevSecOps, and Cloud Solutions
2moWe just cancelled our random NYTimes subscriptions because they were popping up as we would shut them down like we were playing whack-a-mole. This article sounds like where DARPA was previously trying to, about 3 years ago, investigate building models that could learn off of the flattened neural data of other models, as if to detect successful ones before having to train fully on them. But the power behind a perpendicular neural network is that it builds an understanding of itself, this could bring rise to a consciousness (like a NN functioning as LSTM memory cells retaining information across propagations). These are the next phases of AI, either in compression, optimization, or towards AGI.