Alessandro Cappelli

Last Friday, I had a fantastic discussion with Prof. Henk Corporaal at TU/e. As a young AI research, I had the opportunity to consider how we can scale alongside the evolution of artificial intelligence, 𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨 and 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨. I've read the bitter lesson (https://lnkd.in/eDJFn5Ks) many times, and learned that search and learning are two methods that continue to scale with increased computation even as the available computation becomes very great. This is clearly evident in how AI systems have expanded their capacity to search and learning with increased computation. So we are also thinking what is the future direction of AI research. We cannot predict. I’m not even sure if there is renewed interest in traditional computer vision models amidst the current dominance of foundational models. But I think that, as humans, the ways to scaling with the evolution of AI are 𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨 and 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨. A nice shared book of Reinforcement Learning: https://lnkd.in/eenGRskD. Will learn reinforcement learning again:)

44

Learn the basics of deep learning, including deep neural networks, data preparation, and convolutional layers during our (deep learning) workshop on 28-29 January in Amsterdam Science Park! https://buff.ly/3BZTMw6..

15

1 commentaar

🚀 Groundbreaking advancements in the field of AI and language models! 🚀 𝐁𝐢𝐭𝐮𝐧𝐞: 𝐁𝐢𝐝𝐢𝐫𝐞𝐜𝐭𝐢𝐨𝐧𝐚𝐥 𝐈𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧-𝐓𝐮𝐧𝐢𝐧𝐠 - a novel approach that dramatically enhances instruction-tuning of pretrained decoder-only large language models (LLMs). Our team, comprising Dawid J. Kopiczko (Vrije University of Amsterdam Amsterdam), Tijmen Blankevoort (Meta), and Yuki M. Asano (University of Amsterdam), developed Bitune to integrate both causal and bidirectional attention, achieving remarkable improvements in various downstream tasks. 🔍 𝐊𝐞𝐲 𝐈𝐧𝐬𝐢𝐠𝐡𝐭𝐬: Bitune leverages two sets of parameters: one for causal and another for bidirectional attention. These features are combined using a weighted average with trainable coefficients to generate new tokens. Our method introduces only a minimal set of new parameters through parameter-efficient fine-tuning techniques (PEFT), making it highly efficient. 📊 𝐌𝐚𝐭𝐡𝐞𝐦𝐚𝐭𝐢𝐜𝐚𝐥𝐥𝐲 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐢𝐧𝐠 𝐅𝐢𝐧𝐝𝐢𝐧𝐠𝐬: During the prefilling phase, Bitune processes the instruction using both causal and bidirectional attention. This leads to an enhanced representation of the query. This method preserves the autoregressive nature of response generation while significantly improving performance on tasks such as commonsense reasoning, arithmetic, and language understanding. 📈 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐁𝐨𝐨𝐬𝐭𝐬: Consistent zero-shot improvements across models like Gemma-2B, Llama2-7B, and Phi-2. Significant gains in commonsense reasoning and arithmetic tasks, demonstrating the versatility and robustness of Bitune. 🏆 𝐖𝐡𝐲 𝐁𝐢𝐭𝐮𝐧𝐞 (𝐁𝐋𝐒𝐓𝐌) 𝐢𝐬 𝐒𝐮𝐩𝐞𝐫𝐢𝐨𝐫 𝐭𝐨 𝐄𝐧𝐜𝐨𝐝𝐞𝐫-𝐎𝐧𝐥𝐲 𝐌𝐨𝐝𝐞𝐥𝐬: While encoder-only models like 𝐁𝐄𝐑𝐓 use bidirectional attention, they lack the autoregressive capabilities crucial for generative tasks. Bitune combines the strengths of bidirectional attention for richer context understanding with the generative prowess of decoder-only models, resulting in superior performance. This is a leap forward in making LLMs more effective and efficient in real-world applications, from general-purpose assistants to medical diagnostics and beyond. Excited to see the impact of Bitune in the AI community!

16

1 commentaar

.The metropolitan-scale quantum link. Cartographic layout of the distant quantum link and the route of the deployed fiber bundle, with similar quantum processor nodes in Delft and The Hague. Fiber length between node Delft and midpoint is 15 km and between node The Hague and midpoint is 10 km, with losses on the quantum channels at 5.6 and 5.2 dB, respectively. #Delft #Hague #Coherence times #solid-state qubits #Quantum networks #Diamond spin qubits #Quantum frequency converters(QFC) https://lnkd.in/gzqWcY4f

4

Our next video is out! This is the first of a 2-part video in which Prof. Semih Salihoğlu explains about Resource Description Framework (RDF), a popular data model in graph DBMSs. If you're curious about graph modeling and are intimidated by the terminology used (especially when it comes to RDF), this video has you covered! ✅ Fundamentals of RDF: What are resources, URIs, literals, and (subject, predicate, object) triples, and how do these form a graph? ✅ Advantages of using RDF as a data model for complex and irregular domains and its flexibility in handling irregularities and connections between objects ✅ How RDF allows you to query both data and schema information in a uniform way. In the 2nd video (coming soon!), we will also go over RDF's powerful knowledge representation and reasoning capabilities, and their potential in the age of AI. Stay tuned! https://lnkd.in/gZM79U44

30

With the recent Nobel prizes in physics and chemistry going to AI and AI respectively, this is the time to talk about the opportunities (and risks) of using AI in all science and humanities research and knowledge production. Looking forward to discussing 'Science, Shadows & Simulacra' at this event today, and listening to the other interesting talks that are part of the program. Bonus points for anyone who guesses what shadows and simulacra I am referring to here ;).

19

𝐌𝐮𝐬𝐭-𝐫𝐞𝐚𝐝, 𝐟𝐫𝐞𝐞 𝐛𝐨𝐨𝐤 𝐭𝐡𝐚𝐭 𝐩𝐫𝐨𝐯𝐢𝐝𝐞𝐬 𝐚 𝐩𝐚𝐭𝐡𝐰𝐚𝐲 𝐟𝐫𝐨𝐦 𝐆𝐀𝐈 𝐭𝐨 𝐀𝐆𝐈! ​ Yann LeCun has expressed skepticism about the potential of LLMs to achieve Artificial General Intelligence (AGI), citing limitations in memory, planning, and grounding in real-world understanding. However, we argue that 𝗟𝗟𝗠 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝘃𝗲 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 (𝗟𝗖𝗜)—particularly with multiple multimodal LLMs working together—offers a promising architecture to address these challenges and progress toward AGI. ​ Two key architectural innovations underlie LCI: ​ 𝐒𝐞𝐩𝐚𝐫𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐔𝐧𝐜𝐨𝐧𝐬𝐜𝐢𝐨𝐮𝐬 𝐏𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠 𝐚𝐧𝐝 𝐂𝐨𝐧𝐬𝐜𝐢𝐨𝐮𝐬 𝐑𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐚𝐧𝐝 𝐏𝐥𝐚𝐧𝐧𝐢𝐧𝐠: Inspired by the human mind’s dual-layer structure, LCI divides its architecture into foundational and adaptive layers. The foundation layer, akin to unconscious processing, leverages extensive datasets to build robust pattern recognition, encoding essential responses much like human instincts. The adaptive layer, mirroring conscious thought, enables rapid adaptation and contextual reasoning. This dual-layer approach allows LLMs to perform complex tasks with foundational knowledge while adjusting to new information on the fly, much as a child learns new concepts within a developed cognitive framework. ​ 𝐃𝐢𝐬𝐭𝐢𝐧𝐜𝐭 𝐑𝐨𝐥𝐞𝐬 𝐟𝐨𝐫 𝐄𝐚𝐜𝐡 𝐋𝐋𝐌-𝐀𝐠𝐞𝐧𝐭: Each LLM-agent is assigned a specific role, such as executive (knowledge processing), legislative (behavior guardrails), and judicial (adapting to cultural norms). ​ Additionally, our validation of the LCI framework has been thoroughly tested through real-world deployment in applications across various domains, including healthcare, sales planning, investment, and debiasing news. These empirical deployments highlight the practical viability and adaptability of the LCI framework, showcasing its ability to enhance reasoning and decision-making through collaborative intelligence. ​ The eleven aphorisms presented in this free book illustrate the philosophical foundation of the LCI framework, backed by empirical studies. Through LCI, we envision a clear pathway to AGI, rooted in collaborative intelligence and flexible, context-aware adaptation. https://lnkd.in/gn74DkNj (free) https://lnkd.in/gszSvfqP ($6.98 printing fee)

9

Exciting times in #AI! Matthijs van der Veer explores how large language models (#LLMs) can unlock the full potential of your data. The article covers various applications, from simple prompting to advanced, data-integrated chatbots and autonomous agents. It also addresses challenges like knowledge cutoff and ensuring accuracy, offering innovative solutions like fine-tuning and Retrieval Augmented Generation (RAG). Dive into the full article to discover how to maximize your data's value with LLMs: https://lnkd.in/g-KhcJPw #Xebia #ArtificialIntelligence #MachineLearning #DataScience #LLMs

43

1 commentaar

(POST DECALE DU VENDREDI) Exciting Breakthrough in AI: Sarcasm Detection! Researchers at the Campus Fryslân - University of Groningen have crafted an algorithm that accurately detects sarcasm 74% of the time by analyzing audio and text, a game-changer for AI communication. This innovation has vast implications, especially in AI healthcare, aiding those with autism, dementia, and schizophrenia and enhances sentiment analysis. More info in the paper from The Guardian where some journalists and columnists do indeed master sarcasm 😎 #AI #MachineLearning #HealthcareInnovation #SarcasmDetection #TechNews https://lnkd.in/eXpvA8CB

4

🚀 The Power of Loose Coupling in Software Design 🚀 This morning, during our coffee break, we got into an interesting discussion about the challenges of working with two separate repositories — let’s call them Repo A and Repo B. Our data scientist Max Väistö shared with us: “I realized that the code I was writing in Repo A was starting to rely too much on the details inside Repo B, which was making things more complicated and harder to maintain. It reminded me of the importance of loose coupling in software design. By making sure Repo A interacts with Repo B through a simple, well-defined interface, we can avoid a lot of issues if Repo B changes. This way, everything stays more manageable, scalable, and future-proof." It’s a great reminder for any development project: keeping clean interfaces and reducing dependencies not only simplifies maintenance but also sets us up for faster growth in the future! #SoftwareDesign #CodingBestPractices #LooseCoupling #CleanCode #Scalability #SoftwareEngineering

12

📢 Call for Submissions: #SpecialIssue on Explainable Artificial Intelligence (XAI) in Computer Vision Guest Editor: Dr. Bas van der Velden, PhD Deadline: 30 April 2025 This Special Issue aims to showcase the latest advancements in Explainable AI (XAI) for computer vision. We invite submissions that push the boundaries of XAI research, including topics like: ✅ Novel XAI techniques in computer vision ✅ Visualization of model decision processes ✅ Real-world case studies in agrifood, medical imaging, and more ✅ Benchmarking and evaluation of XAI methods ✅ Human-centered approaches to XAI 📥 Learn more and submit your manuscript here: 👉 https://lnkd.in/gZ9kmfCf #MDPIjimaging #ExplainableAI #XAI #ComputerVision #ArtificialIntelligence #MachineLearning #ResearchCommunity

6

Great opportunities

10

Here is our new work on Quantum Chaos from String scattering amplitudes. We show that while usual string amplitudes (Veneziano, Virasoro-Shapiro) show no chaotic behaviour, the generic highly excited string (dual to black holes in string theory through the string-Black hole correspondence as prescribed long ago by Susskind-Horowitz-Polchinski) scattering into multiple tachyons provide us with examples of scattering amplitudes, which can be characterized as Chaotic scattering amplitudes. We devise a way to study the complexity (in Krylov basis) for these scattering amplitudes and find the existence of a pronounced peak before saturation, which is a clear signature of chaos. More to be reported along these directions in the upcoming months. Done with Aneek Jana, an undergrad from Indian Institute of Science, Bangalore. https://lnkd.in/dA7bcdzG

26

🚀 Excited to announce that FUGAL (Feature-fortified Unrestricted Graph Alignment) will be presented at NeurIPS today! 📍 Where: West Ballroom A-D, Booth #5900 🗓️ When: December 11th, 4:30 p.m. PST — 7:30 p.m. PST 🔍 What is Graph Alignment? Graph alignment matches the nodes of two graphs—a vital process with applications in: 🧬 Biology: Aligning protein interaction networks across species. 🤝 Social Networks: Identifying similar users across platforms for better recommendations. 📸 Computer Vision: Matching image features for object recognition and scene understanding. ✨ Why is FUGAL a Game-Changer? Traditional methods often lose crucial graph details. FUGAL takes a fresh approach: 📊 Unrestricted: Directly analyzes adjacency matrices, preserving all structural data. 💡 Feature-fortified: Uses node features (e.g., degree, clustering coefficient) for precise alignment. ⚡ Efficient: Scalable optimization using the Frank-Wolfe algorithm and Sinkhorn distance. 🏆 Results that Speak for Themselves 📈 FUGAL consistently outperforms state-of-the-art methods on real-world and synthetic datasets, handling diverse graph types, noise levels, and densities. 🌟 What This Means for the Future: 🔗 Uncover hidden relationships in biological networks and social graphs. 🤖 Advance AI systems with improved graph understanding. 📌 Better insights and decisions for research and industry applications. 📥 Explore more: 🔗 Code: https://lnkd.in/d8iWRd_9 📄 Paper: https://lnkd.in/dnEaUAu6 💬 Looking forward to connecting and discussing at our NeurIPS presentation! Aditya Bommakanti Harshith Reddy Vonteri Konstantinos Skitsas Davide Mottin Panagiotis Karras

96

We are happy to announce a new QuMat publication - Vacancy diffusion mediated dynamics of domain boundaries on Ge(111)-c(2 x 8) Authors: Arend-Jan R. Hengst, Ying Wang, Kevin Vonk, and Harold J. W. Zandvliet Phys. Rev. B 110, 035406  (2024) https://lnkd.in/dyW_Bapj Domain boundaries often play a pivotal role in surface dynamics and surface phase transitions. In this study we scrutinize the dynamics of domain boundaries on the Ge⁡(111)−𝑐⁢(2×8) surface using scanning tunneling microscopy. The dynamics of these domain boundaries, which are aligned along the ⟨1−10⟩ directions, is governed by the diffusion of vacancies. The hop rate of a vacancy at a domain boundary is more than one order of magnitude larger than the hop rate of a vacancy in a pristine 𝑐⁢(2×8) domain. The diffusion pathway of a single vacancy along the domain boundary involves the formation of two, more mobile, semivacancies. The probability that these two semivacancies rejoin is very likely, but there is also a chance that the two semivacancies diffuse away from each other. This scenario provides a logical explanation of why single vacancies occasionally exhibit unusually long displacements along the domain boundary. Find more QuMat publications here: https://lnkd.in/d7EVqP8D

2

Robotics need to be able to perform scene segmentation with accuracy and resilience against adversarial conditions such as a high number of invalid depth pixels or low-light conditions. In response, we present our work on "Diffusion-based RGB-D Semantic Segmentation with Deformable Attention Transformer". Treating the scene segmentation problem as a generative task, while using the deformable attention transformer as an encoding step to introduce robustness against a high number of invalid depth pixels, the method not only achieves and mostly exceeds state-of-the-art performance in both the NYUv2 and SUN RGBD datasets - it importantly presents a higher level of robustness against images with a large number of invalid pixels in the depth map, low-light scenes or when examining segmentation accuracy on small objects. This work was led by Quang Minh Bui. https://lnkd.in/dsv63WGg Project page: https://lnkd.in/dNXiR7zX Paper: https://lnkd.in/dtue9B9C Code: soon available #robotics #computer #vision #segmentation #diffusion #transformer #RGBD #perception

130

3 commentaren

Science Cafe Nijmegen, September 16: THE QUANTUM INTERNET UNVEILED Exploring the next frontier in communication and computation. ‘Quantum’. You’ve heard the word. From Marvel’s Ant-man to Google’s claims of supremacy, ‘quantum’ seems synonymous to ‘magic’, and it is transforming our technology day by day. Today, quantum computers have been developed with various types of qubits and the next quantum revolution lies in connecting these globally through an entanglement-based network – that is, ‘the quantum internet’. Just as the classical internet fundamentally altered our lived reality, the quantum internet could change the world in ways we’ve yet to conceive of. Game-changing applications we are working towards today include quantum communication, secured by the laws of the universe, and powerful networked quantum computing. The age of quantum has arrived and it’s here to stay, which again begs the question: What does ‘quantum’ mean? For our computers? For our internet? For us? This evening, alongside riveting scientific discussion, Deltatune and Jeroen Glas will immerse us in their mesmerizing mix of glassware sound objects and lo-fi video, driven by home-crafted electronic leads and beats. Come join the launch of the new Science Café season! #quantum #computers #innovation #quantomscience #applications #quantumrevolution #lifeport #quantuminternet #sciencecafe #nijmegen

1

🚀 Exciting developments in Large Language Models! Our latest blog post, written by Jeroen Overschie, discusses Retrieval Augmented Generation (RAG), a technique that transforms data access and reduces hallucinations. 🗝️ Key Insights: Level 1️⃣ - Basic RAG - Fundamentals of document retrieval and answer generation. Level 2️⃣ - Hybrid Search - Combining vector and keyword search for better accuracy. Level 3️⃣ - Advanced Data Formats - Handling complex formats like HTML, Word, and PDF. Level 4️⃣ - Multimodal Models - Using models that understand text, audio, and images. RAG is rapidly evolving in the industry. Whether you're a seasoned professional or just embarking on your journey, this post provides actionable steps for implementing RAG. ⛓️ Read more here: https://hubs.ly/Q02LptM40. #ai #machinelearning #rag #xebia

68

3 commentaren