Jurriaan Schreuder

Groundbreaking AI Decodes Brain Activity into Images 🧠🖼️ Exciting developments in neuroscience and AI! Researchers at Radboud University have created a novel AI system called PAM (Predictive Attention Mechanism) that can reconstruct visual stimuli from brain activity with unprecedented accuracy. Key points: • PAM dynamically identifies relevant neural patterns • Achieves state-of-the-art image reconstructions • Provides insights into visual processing across brain regions Potential applications: • Enhanced brain-computer interfaces • Deeper understanding of visual cognition • Advancements in neuroscientific research This breakthrough could significantly impact fields like neurology, cognitive science, and AI development. Curious to hear your thoughts. How might this technology shape future research and applications? #Neuroscience #ArtificialIntelligence #BrainDecoding For those interested, the full study is available here: https://lnkd.in/etpFuKWB

8

✅️ #Reference van Rooij, I., Guest, O., Adolfi, F. et al. Reclaiming AI as a Theoretical Tool for Cognitive Science. Comput Brain Behav (2024). https://lnkd.in/d8typUgS A few days ago, the #SamAltman prediction of an #ASI "in a few thousand days" was made, but what about the #AGI? The Artificial General Intelligence, such as #AI as smart as the human brain, seems instead in each AI Company announcement to be just around the corner, closer than ever, but in reality, that's questionable. Let me clarify from the beginning why #ThisPaperMatters and you can't miss it: 🗣 "Despite the current hype surrounding "impending" AGI, this practical infeasibility actually fits very well with what we observe (for example, running out of quality training data and the non-human-like performance of AI systems when tested rigorously)." The latest publication from Computational Brain & Behavior by the paper's lead author, Prof. Van Rooij (Radboud University ), pours cold water on the hype and marketing promoting the relentless pursuit of AI. 💧 Frequently in my posts, you have found how that challenge is #overshadowing human inalienable principles and polarizing people against Institutions where regulations are protecting privacy and these principles simply because their work is "slowing down" the diffusion or adoption of the "latest features," outlined for achieving human-like performance in "reasoning" or breaking down concepts to emulate the underlying mechanisms of the #CognitiveScience of our brain. 💡 #What do we really want to achieve? And #why? 🤷 That's not at all a "wow" effect, as with any new paper I share, but to point out and underline how researchers are well aware of the #complexity of creating human-like or human-level artificial intelligence through current ML approaches is practically #infeasible, despite the current hype surrounding AGI. 💡The risk, in my opinion, is to limit or simply reframe human thinking as a #downgrade simply because an AI, according to some #benchmarks, can be "smarter" and "faster" at solving quizzes. So, is our life just a #puzzle? 🧩 ✅️ And more: ➡️ The #Makeism which is the idea that cognition can be understood as a form of computation: known as "#SyntheticMethodology" or "understanding by design and building" it refers to the quote: 🗣"What I cannot create, I do not understand" by physicist Richard Feynman. ➡️ #Intractability ✔️AI-by-Learning is NP-hard ✔️Resource requirements grow exponentially with input size ✔️Even for moderate inputs (e.g., 15-min!), required samples exceed atoms in universe ✔️"Big Data" is insufficient to cover the astronomical space of possible situations ✔️AI companies are running out of usable data ✔️Datasets becoming more homogeneous, not more complex 💡Isn't it perhaps the case to reconsider the role of AI in cognitive science as a tool to understand the natural cognition rather than replicate it? Let's continue in the comments.👇 #Psychology #Cognitive #AI

44

6 commentaren

Closing thoughts on PyData Eindhoven, data scientists have a very good understanding regarding good uses cases for Generative AI, and, as well, use cases that don't really fit Generative AI and are better served by other approaches. For those use cases, the discussion is about monitoring, quality control and management of AI applications. My talk was the first talk in the back-up list, however, none of the titular roster speakers needed to excuse him/her self, so it would be next year. For those wondering what my talk was to be about, check the following repo :) https://lnkd.in/ehwC7h4H

12

In PyData Eindhoven definitely the building of trusted Generative AI applications is the main theme. Not only the building of Retrieval Augmented Generation pipelines is being discussed, but also, how to make them performant and traceable. A lot of the discussions made me remember some talks by Dr. Chris Hillman (Who you should definitely follow), and this article by our colleague Chetan Hirapara (Who you should also definitely follow). https://lnkd.in/g5JYrhHX

32

2 commentaren

🎨 **Dutch National Flag Algorithm Explained!** 🇳🇱 In this 4th reel of our array programming series, we dive into the Dutch National Flag Algorithm. Learn how this powerful technique efficiently sorts arrays with three distinct values. Whether you’re dealing with color sorting or any problem with three categories, this algorithm will streamline your process. Tune in to see how it works and improve your array manipulation skills! 💡📊 #Programming #Algorithm #DutchNationalFlag #ArraySorting #TechTips #Coding

15

"TheaterGen: Character Management with LLM for Consistent Multi-turn Image Generation" Junhao Cheng et al. A Game-Changer in Multi-Turn Image Generation! 🌟 While diffusion models have shown a surprising advancement for high-quality and high-fidelity image generation, multi-turn image generation still faces challenges in semantic and contextual consistency in multi-turn dialogue-driven image generation. Addressing this pressing needs, TheaterGen emerges as a game-changer in the realm of multi-turn image generation, introducing a novel approach that seamlessly integrates large language models (LLMs) and text-to-image (T2I) models. TheaterGen's innovative architecture orchestrates a sophisticated interplay between LLMs and T2I models, akin to a captivating theatrical performance. Acting as the "Screenwriter," LLMs engage in multi-turn interactions, crafting a standardized prompt book that serves as the blueprint for image generation. This meticulous planning ensures: 1) Semantic consistency between images and texts 2) Contextual coherence across multiple interactive turns Moreover, TheaterGen's prowess extends beyond theoretical concepts, as evidenced by its remarkable performance on the CMIGBench (Consistent Multi-turn Image Generation Benchmark). TheaterGen's superiority certaintly shines through, outperforming state-of-the-art methods by significant margins and setting a new benchmark in AI-driven image synthesis. Project page: https://lnkd.in/gxjb6s9E Code repo: https://lnkd.in/gxuPdXkX Arxiv preprint: https://lnkd.in/gkg5VnzE #ImageGeneration #DiffusionModels #LLM #ResearchPaper

1

(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

When invited to the Super Data Science Podcast, Ritchie Vink, author of Polars, mentioned an upcoming Polars book set to launch in January 2025: "Python Polars: The Definitive Guide" by Jeroen Janssens and Thijs Nieuwdorp. The official Github repository already has a lot of code examples that worth checking out: https://lnkd.in/dJGbwQnM The Early Release version can also be read on O'Reilly Learning platform: https://lnkd.in/daTEf8Tv The book is offering guidance for you to:  - Process larger-than-memory datasets at record speed - Apply the eager, lazy, and streaming APIs of Polars and decide when to use which - Transition smoothly from Pandas or Spark to Polars - Integrate Polars into your existing codebase - Work with Arrow and Parquet to efficiently read and write data - Translate complex ETL tasks into efficient and elegant queries I started reading the Early Release version and I am enjoying it so far. I am looking forward to the final version of the book. #Python #DataFrameLibrary #Polars

40

2 commentaren

Yesterday I was wrong and did not notice it. Today I stand corrected! I did not realize that the struggle of AI music to generate consistent song structures (intro, chorus, verse, pre-chorus, bridge, outro) is still a big challenge. It is not. Suno is now generating fairly consistent songs. In other words: We are currently witnessing the sixth-finger phenomenon to leave the stage. I guess it is safe to assume that it is only a matter of months (not years) until it is very hard to tell AI-generated and human-produced music apart. https://lnkd.in/ddyyyq8X

7

6 commentaren

Are open LLM models truly open? 🤔 It seems they might not be. Researchers at Radboud University assessed over 40 large language models and six text-to-image generators, including OpenAI’s #ChatGPT and #DALL·E2. They evaluated 14 characteristics concerning Availability (source code, pretraining data, base weights, fine-tuning data, etc.), Documentation (code, architecture, preprint paper, etc.), and Access (downloadable package and open API). Results: Of the language models, OLMo7B Instruct and LLM360/AmberChat were the only fully available. BigScience’s BLOOMZ was the only model that was fully open in documentation. Some prominent “open” models scored less well. Alibaba Group’s Qwen 1.5, Cohere’s Command R+, and Google’s Gemma-7B Instruct were judged closed or partially open for most characteristics. Neither Meta’s Llama 2 Chat nor Llama 3 Instruct achieved any open marks. Larger corporations often claim openness while providing limited transparency compared to smaller entities that adhere more closely to open-source principles. The study highlights the risks of relying solely on licensing to determine openness, as it can be manipulated to meet minimal standards without genuine transparency. It's time regulatory bodies adopt composite and gradient approaches to assess openness. Companies/Businesses adopting LLM models need to scrutinize and check for detailed documentation to ensure that AI systems labelled as open-source genuinely adhere to open-source principles. How can we ensure the AI systems we rely on are truly transparent and accountable? Share your thoughts in the comments. 👇

10

1 commentaar

Building the Ultimate Nerdland Podcast Chatbot with RAG and LLM: Step-by-Step Guide. Large Language Models (LLMs) have become a hot topic in the tech world. For those of us in Belgium (or the Netherlands) with a passion for technology and science, one of the go-to resources is the monthly podcast "Nerdland," which delves into a variety of subjects from bioscience and space exploration to robotics and artificial intelligence.  Recognizing the wealth of knowledge contained in over 100 episodes of "Nerdland," A simple thought came to our minds: why not develop a chatbot tailored for "Nerdland" enthusiasts? This chatbot would leverage... #techcommunity #azure #microsoft https://lnkd.in/g_TGcqQ9

1

I've danced with algorithms from the dawn of Machine Learning to the psychedelic realms of Generative AI. 💃🤖 From crunching numbers with classic ML to crafting dreams with GenAI, I've seen it all. 🚀 I've wrangled models in the cloud, built MLOps pipelines that would make a DevOps engineer weep with joy, and dived deep into the enchanting world of LLMOps. 🔮✨ With battle scars from countless projects and a brain overflowing with knowledge, I'm here to be your AI sherpa. Need a resume that screams "hire me"? Let's craft it together. Stuck on a project? I've probably faced a similar challenge. Curious about the nitty-gritty of ML or the magic behind LLMs? Let's dive in! 🤿 Your success is my mission. Let's build something extraordinary. 💪 #ML #GenAI #LLMOps #MLOps #Mentorship #AIExpert #CloudExpert #TechGuru #LetsLearnTogether https://lnkd.in/ggcrcPre

16

3 commentaren

Are you building a SaaS-based #startup using #LLM? Here is this tool I found for your #RAG (Link in first comment), a small, portable #vector #database powered by SQLite for on-device RAG:  sqlite-vec is a new vector search SQLite extension written entirely in C with no dependencies, MIT/Apache-2.0 dual licensed. sqlite-vec queries: - 1 million 128-dimensional vectors in just 17ms - 500,000 with 960-dimensional vectors in 41ms sqlite-vec supports: 💾 Matryoshka embedding slicing 💡 Binary quantization reduces storage by 32x with minimal accuracy loss 🤏🏻 L2, cosine and Hamming distance calculations 🧮 Retrieval against #Python List and NP Arrays 🛠️ SDKs for Python, Javascript, Go, Rust, Wasm and more 🧠 local direct embedding with “sqlite-lembed” for gguf models and  Llama.CPP ☁️ remote embedding with “sqlite-rembed” for  @OpenAI  compatible APIs

7

3 commentaren

🚀 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

🤖 xLSTM better than Transformer models? Let's explore this week's Magic AI News together.👇🏽 1️⃣ xLSTM: Language model from Europe The European AI startup NXAI has presented a new architecture for language models in a paper. The new architecture is based on the familiar LSTM architecture. It is called xLSTM (Extended Long Short-Term Memory). For an in-depth look, you can check out the full paper. 👇🏽 https://lnkd.in/eZA3sdZW 2️⃣ Apple invests in AI infrastructure Apple plans to integrate its own M-chips into servers. Generative AI requires big server parks with much computing power. 3️⃣ DeepSeek unveils a new open-source LLM DeepSeek's unveiling of DeepSeek-V2, an open-source MoE large language model, shows impressive performance with fewer parameters. It's a powerful model, especially for tasks involving the Chinese language and coding. 🎬 Want to know more about these exciting developments in AI and the tech world? You can find our thoughts on the news in the full newsletter! 👇🏽 #artificialintelligence https://lnkd.in/egMJPNCr

9

2 commentaren

We are still very early in machine learning for protein engineering. A core belief we have held at Cradle is that the only way to build a high quality product with best in class generative models is to invest in a low-latency, high throughput lab. This lab allows us to build 'foundational datasets', a/b test new releases of our models and demonstrate generalisation across many protein modalities (i.e. enzymes, antibodies, car-t's, peptides etc). Two years ago we set a high bar: we wanted our lab to be able to complete an experimental round in 2 weeks, at 384 throughput for many different assays. This meant trying to reduce latency in every step of our DBTL cycle. In this blog series we share the methods we developed. This month: express and purify 384 variants in just 3 days of relatively light lab work and not breaking the bank. The protocol is added to the blog-post. Give it a try and feel feel to ask us questions! Follow Cradle for more methods being published to speed up your research and development. The blogpost with protocol link is in the comments!

222

10 commentaren

🌟 Master the Future with Generative AI: A Step-by-Step Guide to Implementation 🚀 Generative AI is reshaping industries by unlocking creativity and enabling efficiency. From crafting novel content to enhancing workflows through data augmentation, this is your go-to guide for a seamless AI journey! ✨ What’s inside this infographic? 🔍 Stage 1: Define the Generative Task Are you focusing on creative outputs like content generation or functional tasks like dataset augmentation? Clear goals lead to better results. Identify data requirements, set quality metrics like diversity and coherence, and start strong. 🧹 Stage 2: Data Preparation & Conditioning Your AI model is only as good as your data. Data Cleaning ensures accuracy and consistency. Data Conditioning aligns data formats with your model’s needs. Data Augmentation expands datasets to improve robustness. 💡 Stage 3: Model Selection & Architecture Choosing the right model is key: Pre-trained models like GANs or Transformers can save time. Customizing models for your task enhances performance. Fine-tuning hyperparameters ensures optimization for your data. ⚙️ Stage 4: Training & Evaluation Train your model effectively: Build the right training setup (hardware, software, and data). Use appropriate loss functions to guide learning. Apply evaluation metrics like FID, BLEU, or KID for accuracy. Iterative refinement helps tackle overfitting and improves outputs. 🔧 Stage 5: Post-Training Refinement Post-training is where the magic happens: Fine-tune your model for specific use cases. Leverage prompt engineering to get the best results. Address biases to ensure ethical and fair AI implementation. 🚀 Stage 6: Deployment & Monitoring The finish line is just the beginning: Seamless Integration into workflows ensures adoption. Performance Monitoring keeps your AI effective in real-world scenarios. Enable continuous learning to keep the model evolving with new data. ✨ Key Takeaways Data Quality is non-negotiable. High-quality, diverse data drives success. Balance model complexity with resource efficiency. Ethics & Transparency should be at the core of your implementation. Focus on user experience to ensure adoption and impact. 🔥 Why this matters: Generative AI is no longer a buzzword – it’s a game-changer. But success lies in execution. Start now to stay ahead of the curve and lead the change! 💬 What excites you the most about Generative AI? Share your thoughts in the comments or reach out to discuss how we can leverage AI for innovation together. #GenerativeAI #ArtificialIntelligence #Innovation #FutureOfWork #MachineLearning #DeepLearning #DataScience

7