Post de Thibaut de la Grand'rive

📢 NEWS: The 2023 State Of Open Data report is here - published by Digital Science, figshare and Springer Nature Group! 😃🙌 #StateOfOpenData Now in its 8th year, the 2023 State Of Open Data report is based on responses to a global survey from more than 6,000 researchers, sharing their attitudes towards and experiences of open data. Find out more about the 2023 report: https://ow.ly/G3oE50Q7l68 See our joint press announcement: https://ow.ly/2uL750Q7l6a Register now for our first webinar about the report - The Headlines: https://ow.ly/9oMo50Q7l66 This year’s State of Open Data report has five key findings: ⚡ Support is not making its way to those who need it. In 2023, researchers overwhelmingly are still not getting the support they need to share data openly, with three quarters of respondents saying they have never received support with sharing data. ⚡ One size doesn't fit all - for academic disciplines & geography. A more nuanced approach is needed. ⚡ Challenging stereotypes. Career stage is not a significant factor in open data awareness or support levels. ⚡ Credit is an ongoing issue. This is a big one! For eight consecutive years, our survey has revealed this recurring concern among researchers. ⚡ AI awareness hasn’t translated to action. For the first time, we asked if researchers are using #AI tools for data collection, processing and metadata creation. Find out more about the 2023 State of Open Data report here: https://ow.ly/G3oE50Q7l68 The full report can be accessed on Figshare: https://ow.ly/k5qI50Q7l67 #OpenData #OpenResearch #OpenScience #ArtificialIntelligence #data #DataSharing

🔍 Exploring Fruits Dataset for Classification 🍏🍊🍓 📁 Dataset Overview: The dataset consists of images of various fruits, categorized into six classes: fresh peaches, fresh pomegranates, fresh strawberries, rotten peaches, rotten pomegranates, and rotten strawberries. 📊 Data Exploration: Upon examining the dataset, I discovered a slight imbalance in the distribution of images across classes. To address this, I utilized oversampling techniques to balance the dataset, ensuring equal representation of each class. 🖼️ Data Augmentation: To enhance the diversity of the dataset and improve the model's robustness, I applied various data augmentation techniques such as rotation, shifting, shearing, zooming, and flipping to the images. This helped in generating additional training samples, thereby enriching the dataset. 🛠️ Model Architecture: For the classification task, I leveraged the powerful VGG16 convolutional neural network architecture pretrained on ImageNet. By fine-tuning the model and adding custom fully connected layers, I tailored it to suit the specific requirements of the fruit classification task. 🔢 Training and Evaluation: I split the dataset into training and testing sets and trained the model using the augmented data. During training, I monitored key metrics such as accuracy and loss to evaluate the model's performance. The model achieved a satisfactory accuracy of [insert accuracy value] on the test set, demonstrating its effectiveness in classifying fruit images. 📈 Next Steps: Moving forward, I plan to further optimize the model by experimenting with hyperparameters, exploring different architectures, and fine-tuning the data augmentation techniques. Additionally, I aim to deploy the trained model to classify fruits in real-world scenarios, contributing to advancements in agricultural technology and food quality assessment. Stay tuned for more updates on this exciting project! 🚀 kaggle:- https://lnkd.in/et_UuDwk UpVote 🔼 #MachineLearning #ComputerVision #DataScience #FruitClassification #DeepLearning #AI #ArtificialIntelligence

ai.plainenglish.io: The text explains different approaches for handling missing values in categorical variables. These approaches include removing rows with missing values, imputing with the mode or a new category, and training a model to predict the category. The author suggests using a probabilistic approach based on the distribution of available values. They provide an implementation using Python code and discuss the importance of sample size and the possibility of adding noise to the distribution. The text concludes with a mention of a hybrid approach and encourages readers to follow the writer and explore related content. - Artificial Intelligence topics! #ai #artificialintelligence #intelligenzaartificiale

Explaining Multiclass Classification in ML to kid 🌳🐘 Kid: What's multiclass classification in ML? 🤔 Me: Imagine you're in charge of a wildlife sanctuary, and your task is to assign each animal to its most suitable habitat. 🏞️ Kid: So, it's like ensuring every animal finds its perfect spot among all the available choices! 🐅 Me: Precisely! Kid: Then, how is it used in ML? Me: In ML, we provide the computer with data about various animal features—such as size, habitat preferences, diet etc. & ask it to find the most appropriate habitat zone among various possibilities. Kid: That's intriguing! But, why is it called "1 vs ALL"? 👨🏫 Me: The "1 vs ALL" concept refers to the approach of comparing each animal (the "1") to all other possible zones (the "ALL") to find its most suitable habitat. For instance, when considering a zebra, we compare it against all habitat zones of lion, elephant, and others—to determine the best match. Kid: Does the computer ever mess up, like putting a giraffe 🦒 in the penguin zone? 🐧 Me: Oh, absolutely! Sometimes it's like inviting a giraffe to a black-tie event for penguins—a total mismatch! 🤣 But we're schooling the computer to distinguish between tall orders and chill vibes. I share kid-friendly resources to explain #datascience concepts. You can find all the posted resources here - https://lnkd.in/dNcu-qxb Follow Aditya Chourasiya for more such content! #machinelearning #deeplearning #data #analytics #ai #jobs #nlp #India #bigdata #ml #iot #powerbi #automation #oilandgas #connections #dataanalytics #engineering #computervision #sql

Explaining Multiclass vs Multilabel Classification in ML to kid 🍪 Kid: What's the difference between multiclass and multilabel classification in ML? 🤔 Me: Imagine we're baking cookies, but some aren't just chocolate chip or oatmeal—they're mix-and-match cookies with different flavors and shapes! 👩🍳 Kid: Oh, cookie adventures! So, what's multiclass in cookie land? 😄 Me: Multiclass is like sorting cookies into distinct jars—chocolate chip goes into one, oatmeal in another, and so on. Each cookie gets its own special jar! Kid: Got it! So, what's multilabel then? 🤔 Me: Picture this: a cookie can be both lemon-flavored 🍋 and heart-shaped ❤️ or maybe chocolate-flavored and star-shaped 🌟— like giving each cookie multiple cool labels! Kid: Like superhero cookies with superpowers! 🦸♂️ Me: Exactly! Multilabel lets cookies have many labels, unlike multiclass where each cookie fits into just one jar. 🏷️ Kid: That's genius! Can we make superhero capes for cookies? Imagine them soaring with flavor and data strength! 💪 Me: Totally! Just watch out, these cookies might analyze the best way to beat the cookie jar security system! 🔒 #learnings Multiclass Classification: Assigns data to singular distinct classes, each item fits into a specific category. Multilabel Classification: Allows data to have multiple labels simultaneously, accommodating various attributes per item. I share kid-friendly resources to explain #datascience concepts. You can find all the posted resources here - https://lnkd.in/dNcu-qxb Follow Aditya Chourasiya for more such content! #machinelearning #deeplearning #data #analytics #ai #jobs #nlp #India #bigdata #ml #iot #powerbi #automation #oilandgas #connections #dataanalytics #engineering #computervision #sql #llm

[NEW PUBLICATION] Academics are often criticized for speaking in abstract terms and being detached from the real world. Sometimes this is really true, but not with this paper. Google Trends are a very handy tool for anyone interested in at least tracking the popularity of a particular phenomenon or constructing a sentiment indicator, if not in forecasting any type of economic activity. If you scratch below the surface of machine learning models and data analytics, our new paper in TECHNOLOGY IN SOCIETY (IF 9.2, Q1 WoS) provides some insights for anybody interested in using Google Trends analysis. Key takeaways: 📌 Categorical search is probably a better idea than the strategy of picking specific search terms. 📌 Composite indicators should behave better than single keyword or single category indicators. XGBoost method is particularly useful in that sense. 📌 Data transformations (structural break adjustments, filtering, common trend extraction) may considerably influence the accuracy of your results. Great work, Ivana Lolic and Marina Matošec! Chapeau!