Alena Seredko’s Post

In an exciting development, researchers will be able to identify whether scholarly content has been made available to them (via institution subscriptions or open access) no matter where they are on the web when using the #GetFTR Browser Extension. It will indicate whether content has been retracted or updated at the point of discovery, thanks to GetFTR's collaboration with Crossref and Retraction Watch. Dianne Benham, MBA said “This feature extends GetFTR’s reach to all websites. We hope this will vastly improve the customer experience, elevate the value of our publisher’s offerings, and contribute to Research Integrity” See an example of how GetFTR indicators appear on Wikipedia, the Free Encyclopedia in the comments or read about it on our website: https://lnkd.in/etu_FkCC #openaccess #scholcomms #researchintegrity

We just open sourced our daily discovery algorithm at TIDAL!!! As a part of an effort to become more open and transparent, we're open sourcing how we create our personalized daily discovery algorithm. We also wrote a blog post explaining all the system components, technology stack, machine learning utilization and more. Personally I am extremely grateful for this contribution opportunity. Even though educational resources and datasets for recommender systems are numerous and readily available, I always struggled with finding high quality content explaining real industry experience, know-hows and rules of thumb, as well as seeing how actual production code is structured to scale across teams and projects, all in real life, non controlled environment scenarios, and I hope this can serve as a small reference point to paint an image for how companies build their machine learning and big data systems to create products that users love! Check out the source code at: https://lnkd.in/dG5maFPY And the blog post at: https://lnkd.in/dp5DCvcz Looking forward to your comments, issues and PRs

This project is the graveyard of best intentions: everyone wants to do it, no one can get it right, picking such project is like picking a thumbstone for thyself. What's amazing is absolute beaty of first principles of the proposed approach: from perfecting enginiering foundation until it's all logged 1:1 to perfecting theoretical formulation until it works. No heuristics, just theory from first principles. Does not work? Then theory is wrong. What we missed? And this is type of project where ML does not work. No evaluation off-the-shelff, how even to navigate if what you added is working before deploying? Unimaginabily hard. Yet we solved it. Kudos to Olivier Jeunen who cut through the last mile, which is always the hardest job to do in such projects, especially when it comes to discovering theoretical holes in my implementation, which ultimately made the puzzle to assemble :p

How will @microsoft-power-platform next-gen AI features for #PowerVirtualAgents and #AIBuilder drive a more intuitive experience for citizen developers and accelerate efficiency for professional developers? Find out 👇

Female representation in model training in ML and Deep Learning is only in it's rookie state. What can we do to change that? Here is an example from DeepFaceLab https://lnkd.in/eAqiu4qP Looking at the Readme and how it is presented to me, I wonder if the model can properly create deep fakes where female presenting people are represented. Looking at this and thinking: Are other models trained and presented like this as well? Also the community looks quite female under-presented. Or how can we explain the latest community meetups in #Berlin? Claus Höfele Any plans to improve that? Do you need help?

I'm super excited to return as a speaker to the Haystack US 2024 Search Relevance Conference by OpenSource Connections! Last year's was one of my all-time favorite conferences due to the smart, welcoming community and cozy venue (a small movie theater in Charlottesville VA). My talk will focus on what causes so many ongoing search+AI projects to fail in enterprises, ranging from technology choices to expectation setting and client management, and how to avoid these pitfalls. This project focus is a step higher in abstraction than my talk last year, which focused on technical failure modes of LLMs when combined with information retrieval. Check out my talk abstract here: https://lnkd.in/gDEnhaND and attend the conference if you can! It'll be an eye-opening yet relaxing event with longtime practitioners of enterprise search, perfect for anybody using search+RAG/LLMs/GenAI in industry (So: everybody). #search #llms #rag #genai

Here is my most read article about Emerging landscape of graph related technologies (more than 25 000 Views)! And you can access it from its QR code, if the technology I just start to test works well. I will share the return on experience on the value it can bring, in combination with content publishing througth LinkedIn. Let's follow #emerging_landscape_of_graph_related_technologies #QRCode_technologies

It's been a busy week. I've spoken at two conferences in the past two days! But the fun is not over, I hope you can join me on May 8th as I'll be presenting at the Minnesota Technology Association annual conference, TECH CONNECT! The topic this time will be, Open Source Large Language Models. Through this session and my deep experience, I'll guide you through a framework to help you decide if open-source or closed-source LLMs are the right choice and more importantly, why. Open Source LLMs do and can work for your organization, you need to plan and understand the pros and cons. As always, I'm no one to talk theory, I'm all about the applications of AI, so expect us to put some of these models through the paces through models hosted on Hugging Face. Hope to see you then! https://lnkd.in/gMENRzyP Linda Cummings Lab651 | Custom Software Development & Process Recursive Awesome Josh Braaten Nick Roseth Christopher Ebbert David Espindola Dan McCreary Matthew Versaggi Michael A. Tim Bornholdt Emily McCarthy Allison Widmark #appleidai #applications #artificialintelligence #llm #opensource

In my latest blog post, "Beyond External Embeddings: Integrating User Histories for Enhanced Recommendations," I delve into the realm of external user histories and their integration into recommender systems, extending beyond my previous exploration of external embeddings. The article navigates through cross-domain recommendations, starting with simpler scenarios where external items intersect or connect with internal ones, and transitioning into more complex cases. A key focus is placed on the utilization of two-tower networks, demonstrating their effectiveness in these contexts. Additionally, the post discusses various methods of employing models with external history, including features, candidates, product rules, and more advanced techniques. I strongly advocate for initially focusing on features, highlighting their foundational role in the effectiveness of recommender systems.

Leading companies in the open text to image community have banded together to form the Open Model Initiative in a F--- U to Stability AI's new restrictive license for SD3. Invoke and ComfyOrg two of the most popular interfaces for open source image generation and CivitAI is one of the largest communities sharing fine tuned models. This is basically a revolt by the community who use stable diffusion models. The new org - will be funded by the member organisations and won't take VC money - will produce models that are truly open source - will be respectful of artists and privacy by developing a base model without the ability to generate images based off unconsented artist name or generating likeness of unconsented individuals - will ship something in the next 3 to 6 months (?) That's a lot of wills and it's easier to make an announcement than deliver value, but good on them! I'm curious what people think about the 'respectful of artists and privacy' bit - game changer or more noise? Invoke: https://lnkd.in/gUbFvDsZ CivitAI: https://lnkd.in/g6_Mhmhv