<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Syncmind Lab on Medium]]></title>
        <description><![CDATA[Stories by Syncmind Lab on Medium]]></description>
        <link>https://medium.com/@syncmind?source=rss-a1f822fceef8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*xGQyJazJ2bpwUaYwIKTJ8A.png</url>
            <title>Stories by Syncmind Lab on Medium</title>
            <link>https://medium.com/@syncmind?source=rss-a1f822fceef8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 18:48:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@syncmind/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[MindNet: analysis of the network structure of human brain connectomes.]]></title>
            <link>https://syncmind.medium.com/mindnet-analysis-of-the-network-structure-of-human-brain-connectomes-e9a3af23ecfb?source=rss-a1f822fceef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9a3af23ecfb</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[biotechnology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[healthtech]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Syncmind Lab]]></dc:creator>
            <pubDate>Wed, 03 Apr 2024 19:54:07 GMT</pubDate>
            <atom:updated>2024-04-03T19:54:07.295Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7jtcITHs0_BfRvoX9gCTTA.png" /></figure><p>The utilization of machine learning techniques for analyzing functional magnetic resonance imaging (fMRI) brain scans faces challenges due to intricate data preprocessing steps and uncertainties in hyperparameter selection for constructing graph-based datasets. To tackle this issue, MindNet, a Python-based software package, is under development. It leverages key neural network methodologies to train models, assess model performance metrics, and offer a visual interface for comparing graph neural network (GNN) methods in fMRI analysis. This initiative aims to streamline and standardize brain functional connectivity analysis workflows, conduct experiments across diverse datasets, and enhance research methodologies to yield valuable insights.</p><p>Functional brain networks exhibit intricate topologies portrayed as graphs with distinct regions and connections, amenable to analysis through graph neural networks. GNNs demonstrate potential in addressing various tasks like developing recommendation systems, data analysis, and prediction. This framework establishes a mathematical basis for exploring successful neural network architectures such as CNNs, RNNs, GNNs, and Transformers, enabling the incorporation of structural graph knowledge into neural designs and fostering the exploration of novel architectures.</p><p>fMRI technology, reliant on Blood-Oxygen-Level-Dependent (BOLD) signal registration, is extensively used to identify brain activity zones in both normal and diverse central nervous system disorders, serving as a primary data source for functional brain connectivity analysis. Computational fMRI analysis employs multiple deep learning methods to scrutinize complex relationships between brain regions based on BOLD signals. Contemporary deep learning-driven BOLD signal analysis techniques, including CNNs, RNNs, GNN models, and transformer architectures with attention mechanisms, have exhibited promising outcomes. Transformer-based models have garnered considerable attention for their superior performance in various tasks like time series forecasting, natural language processing, and computer vision.</p><p>The main focus in fMRI data analysis is ensuring the data is well-prepared to conduct effective computational experiments. Training machine learning models with fMRI data necessitates a substantial amount of data due to the limited representation of brain function in each scan. The MindNet facilitates the production of meaningful empirical findings and the formulation of concepts for future exploration in this advancing domain</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e9a3af23ecfb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Federated learning for clinical purposes in simple terms]]></title>
            <link>https://syncmind.medium.com/federated-learning-for-clinical-purposes-in-simple-terms-b0a83528213a?source=rss-a1f822fceef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0a83528213a</guid>
            <category><![CDATA[medtech]]></category>
            <category><![CDATA[federated-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Syncmind Lab]]></dc:creator>
            <pubDate>Mon, 18 Mar 2024 10:17:19 GMT</pubDate>
            <atom:updated>2024-03-18T10:17:19.038Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0wMqjAOGmqw4xbr72tXtCg.png" /><figcaption>The picture above illustrates the federated learning cycle. The model is trained locally on the device (A). Updated models from all devices are aggregated at point B, the global model is updated, and at step C, it is sent back to the local devices.</figcaption></figure><p>The field of machine learning is evolving at an incredible pace and is being applied in many areas of life. The amount of data that needs to be analyzed is increasing. Questions arise: how to efficiently train machine learning models and where to store so much data? And also one of the main challenges is the confidentiality of patient data and compliance with GDPR and HIPAA regulations</p><p>One way to solve such problems is through parallelizing training and data storage. This is precisely the idea behind federated learning, also known as collaborative or “joint” learning.</p><p>Federated learning was first introduced by Google in 2017 in the Google AI Blog “Federated Learning: Collaborative Machine Learning without Centralized Training Data,” alongside the seminal research paper that laid the foundation for federated learning with an in-depth discussion of the new approach called federated optimization, titled “Federated Optimization: Distributed Machine Learning for On-Device Intelligence.”</p><h3>The difference between federated learning and the traditional approach</h3><p>The traditional machine learning pipeline utilizes a central server (local or cloud-based) where the trained model resides. The downside of this architecture is that all data collected by local devices and sensors is sent to the central server for processing and then returned to the devices. This round-trip limits the model’s ability to learn in real-time.<br>In contrast, federated learning is an approach that follows the following cycle: the current model is downloaded from the server to the device, retrained on the same device using local data, and then the updated model is sent back to the central server, where it is aggregated with other models by averaging the weights. Then the improved global model is sent back to the devices.</p><p>Moreover, clients participating in federated learning may be unreliable as they are subject to more failures or dropouts, relying typically on less powerful communication means (e.g., Wi-Fi) and battery-powered systems (e.g., smartphones and IoT devices) compared to distributed learning, where nodes usually represent data processing centers equipped with powerful computational capabilities and interconnected by fast networks.</p><h3>Benefits of Federated Learning:</h3><p>• Federated learning enables devices such as mobile phones to collaboratively train a global machine learning model while keeping the data on the device itself instead of transmitting it to a central server.<br>• Federated learning makes it possible to train the model on peripheral devices such as smartphones, tablets, and IoT devices.<br>• Another important advantage is the preservation of personal data on the device and ensuring the confidentiality of the information used. This ensures data security in various fields, including defense, telecommunications, medicine, etc.<br>• Personalization of the machine learning model.<br>• Federated learning enables real-time prediction since training occurs on the device itself, eliminating the time delay caused by transmitting data to the server and back.<br>• Since the models are on the device, the prediction process works even without an internet connection.<br>• Federated learning reduces the amount of required hardware infrastructure since the used power is the device’s own power, which is sufficient to support model training and prediction.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0a83528213a" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>