Robin Fay’s Post

"In the few-shot function, concrete examples help point the model to the types of information and patterns it should pay attention to." Ilia Teimouri PhD explains how few-shot learning can be useful in text-classification tasks.

Hello, friends! Today was truly special and exciting as we started the day with a Machine Learning course taught by Mr. Ahmed, one of the employees at PDO Company. The lesson covered some crucial topics that I'd like to share with you: 1. The meaning of Machine Learning: It is a branch of Artificial Intelligence that focuses on developing systems capable of learning and adapting automatically from data and previous experiences. 2. The difference between Machine Learning and Software: Traditional software relies on fixed programming, whereas Machine Learning can improve its performance over time. 3. Types of Machine Learning: We discussed three main categories:   - Supervised Learning: The model is provided with labeled training data to make accurate decisions in the future.   - Unsupervised Learning: The model is given unlabeled data and learns from it without external guidance.   - Semi-Supervised Learning: A balance between supervised and unsupervised learning, where the model receives a small amount of labeled data and a large amount of unlabeled data. We couldn't help but feel excited to continue our learning journey and explore more about Machine Learning. Many thanks to Mr. Ahmed for his efforts in explaining these fascinating concepts. Sultan AL-Yahyai Code Academy - OMAN Ahmed Azri #MachineLearning

We added a new initiative over the holidays. "The Learning Journal Initiative" on Medium. 250 Followers in a week. Fifth story just published on "Unstructured Learning Journals" 👇 The link below is a 'friend link' which means anyone can read this article - you don't have to be a Medium member. "The Messy (and Powerful) Art of Unstructured Learning Journals for Thinkers and Writers" YOUR INCUBATOR AND CANVAS FOR IDEAS, THOUGHTS, AND CREATIVITY https://lnkd.in/eHAqCPa4

A really interesting publication on SSL (Self supervised learning) that delves into the rapidly evolving field. The paper summarises each major category of SSL methods and components of each method. Highly recommended if you’re interested in upskilling on SSL: https://lnkd.in/ezyhRf9H #artificialintelligence #ai #machinelearning #Selfsupervisedlearning

Check out this nice paper by my PhD student Christoph Schmidl, as so often in great collaboration with Thiago D. Simão. TL;DR: Yes, reinforcement learning can be used for (job shop) scheduling, and yes, it is important to think about robustness and generalizability! This result was obtained within our MasCot project SAM-FMS, a program by NWO (Dutch Research Council) and TNO-ESI.

Today marks the last day here at the ICAART 2024 conference in Rome. I'm happy that we contributed the paper "A Supervised Learning Approach to Robust Reinforcement Learning for Job Shop Scheduling", which is joint work with Thiago D. Simão and Nils Jansen. Suppose you are facing a complex Job Shop Scheduling Problem (JSSP) instance, where a set of jobs, each consisting of a sequence of operations, must be mapped to starting times for specific machines while minimizing the objective of the total time taken (makespan) and adhering to machine and task constraints. We can use Reinforcement Learning (RL) to learn a policy that acts as a scheduler for the given instance and can compete with other heuristics. However, permutations in job orders of the given instance represent a problem for the policy and lead to a longer makespan, although the resulting schedule should be the same as the one without permutations. Therefore, the RL policy is not permutation invariant and is limited in generalization capabilities. In our recent work, we approach this problem using a three-step method that combines RL and supervised learning techniques. The initially trained RL policy acts as a labeling oracle that generates state-action pairs, which are then augmented with varying permutation percentages to transpose job orders. These state-action pairs serve as data sets for re-training models in a supervised learning setup that uses Dropout layers to improve robustness. If you are interested in more details, feel free to have a look at our ICAART paper! Paper: https://lnkd.in/ec9fBUCs Code: https://lnkd.in/ectvTThu