Frank La Vigne

As systems with single-node ephemeral engines over #S3 become mainstream (hello bauplan!), a new wave of "format aware" caching tools is inevitable. Our friends at University of Wisconsin-Madison and InfluxData just released LiquidCache, a cloud-native cache built on Apache DataFusion, a disaggregated cache based on logical data representation with incredible performances: go and add a star to Xiangpeng Hao repo now! While bauplan cache works without custom formats (it's specifically built to be native for Apache Iceberg and Apache Arrow), it shares many of the same ideas: extracting #S3 scans from queries, re-using entries on overlapping requests and more importantly, representing scans with *logical* filters, supporting relational algebra inference across languages. Our paper from last year with Ryan Curtin is available here (and perhaps some new papers are brewing as well, who knows 😎): https://lnkd.in/eCmedBgf! Congrats to our friends Xiangpeng Hao, Andrew Lamb and Remzi Arpaci-Dusseau for the incredible work: look forward to geeking out on caching the next time we all chat! #iceberg #s3 #arrow #cache

282

5 Comments

Voice AI is quietly hitting an inflection point 🔊 NVIDIA just published a great deep dive on scaling speech recognition for real-world voice agents - latency, streaming accuracy, robustness, and why “ASR quality” isn’t just about WER anymore. 👉 https://lnkd.in/enDUKHvG A few things that stood out: - Streaming + low latency is non-negotiable if you want natural conversations - ASR needs to handle interruptions, background noise, accents, and partial utterances, not clean demo audio - Scaling voice agents isn’t just model quality, it’s systems engineering If you want to actually try building or running voice agents instead of just reading about them, Pipecat is a great place to start: 👉 https://pipecat.ai This is the kind of progress that makes voice agents finally feel… practical. Not demos. Not prototypes. Production. Curious to see where the ecosystem lands this year 👀

146

10 Comments

Mitigating over-smoothing in Graph Neural Networks for node classification through Adaptive Early Embedding and Biased DropEdge procedures By Fateme Hoseinnia, Mehdi Ghatee, Mostafa Haghir Chehreghani has been accepted in “Knowledge-Based Systems“. Highlights: * Proposes Adaptive Early Embedding for nodes to mitigate GNNs over-smoothing. * All nodes need not process the embedding until the last layer for node embedding. * Utilizes Biased DropEdge to reduce over-smoothing convergence speed. * Applies two plug-and-play procedures for deep, multi-layer stacking message-passing GNNs. * Proves theoretically that the proposed model mitigates over-smoothing. Abstract: Graph Neural Networks (GNNs) are widely used for tasks involving graph-structured data across various fields, including computer vision, biology, social media, and traffic prediction. Despite their substantial success, increasing the depth of GNNs can impair the discriminability of node representations, leading to a decline in performance for node classification tasks. This challenge is partly attributed to a phenomenon known as over-smoothing. This paper introduces an Adaptive Early Embedding (AEE) procedure between Graph Convolutional Network (GCN) layers. This method adaptively halts the aggregation of neighboring nodes before the final layer of the main network. By reducing the over-smoothing of node embeddings, we enhance the distinguishability of the data. Another important contribution of this work is using the inter-class Biased DropEdge (BDE) procedure, which effectively propagates beneficial information. The proposed model based on AEE+BDE can be integrated with baseline message-passing GNN models to mitigate over-smoothing challenges. Our experiments show that the proposed model outperforms baseline models. Additionally, we provide theoretical evidence supporting the effectiveness of the AEE and BDE procedures for node classification tasks. Ref: https://lnkd.in/d_BvHDDg The full version of the paper “Mitigating over-smoothing in Graph Neural Networks for node classification through Adaptive Early Embedding and Biased DropEdge procedures” by Fateme Hoseinnia, Mehdi Ghatee, Mostafa Haghir Chehreghani is now available online in “Knowledge-Based Systems“. It can be downloaded by clicking the following Share Link before July 03, 2025. No sign up, registration, or fees are required. https://lnkd.in/dgcBTqZc

127

1 Comment

Our latest post explores why portable memory for AI would be good not only for users but also for application developers, and even the major model labs. Consistent with our idea that open protocols create an “architecture of participation” that enables larger and more robust markets, we provide aprovocation that might lead to such an architecture for AI.

74

5 Comments

SLMs will foster agentic #AI Small Language Models (SLMs) are models that fit on consumer hardware and run with low latency. Models like like Phi-3, Nemotron-H, and SmolLM2 have already matched or outperformed older LLMs on tool use, reasoning, and instruction following. Good paper from NVIDIA and al. Link here : https://lnkd.in/eXSiirEb

88

10 Comments

AI Agents: The Future of Autonomous Systems – Quick Guide  AI Agents are no longer just about chat — they're about autonomy, reasoning, planning, and action. Whether you're building tools or exploring AI workflows, here’s a compact guide to key concepts, tools, and terminology shaping this space: 🧠 Knowledge & Memory AI agents need memory to function intelligently. Tools like: LlamaIndex for document retrieval Pinecone / Chroma for vector memory Redis / Postgres for long-term storage FAISS for similarity search 📐 Core Concepts Plannig: Breaking down tasks into steps Autonomy: Acting without human input Tool Invocation: Using APIs/tools dynamically Reflection: Learning from past actions Statefulness: Remembering context Prompt Engineering: Still crucial 🛠 Popular Tools & Frameworks AutoGen: Microsoft’s multi-agent framework LangGraph: For building stateful workflows LangChain: Modular, tool-friendly agents CrewAI / HuggingGPT / Haystack: Multi-agent, task-based systems 📘 Important Agent Terms ReAct (Reason + Act) Chain of Thought (CoT) Planner–Executor Pattern Agent Loop: Plan → Act → Learn Role-Based Agents Memory Types: Short, long, episodic AI agents are transforming how we automate, reason, and build intelligent systems. From product assistants to research bots, this is where the future is headed. Are you building with agents? Let’s connect and share insights.

240

30 Comments

Read more on my latest post on the Scale AI blog! https://lnkd.in/gZpmwhu4 1.5M+ autonomous AI agents are now interacting with each other online. No humans. Fewer prompts. This is one of the first real examples of AI collectives operating at scale, and it breaks most of our existing risk models.

37

3 Comments

Just read the MIT Technology Review piece on Google releasing energy data for Gemini. It’s interesting to finally see concrete numbers behind what an AI prompt actually consumes, rather than broad estimates or assumptions. As AI use scales, this kind of transparency feels important for understanding the real infrastructure and sustainability tradeoffs behind everyday interactions. Worth a quick read if you’re thinking about where large-scale AI is headed. #AI #AISustainability #AIInfrastructure

13

Florida is protecting and restoring the Everglades with Google AI Google Cloud Results ✅ End-to-end development and deployment of the new tool within 10 months ✅ New interactive, user-friendly interface for both scientists and administrators ✅ Expanded access to hydroperiod data for advanced inundation modeling and environmental monitoring ✅ Data processing times reduced from hours to minutes https://lnkd.in/grd9VR2j #GoogleAI #GoogleCloud #ResponsibleAI #PublicSector

25

From UI frameworks to data dashboards to cutting-edge NLP, at New Math Data, we’ve been busy exploring the cutting-edge technology that moves teams and organizations forward. ✅ Angular 20 is here, and Sriprasad's article touts it as the most developer-friendly release yet. With standalone components, zoneless change detection, and a smarter build pipeline, this version doesn’t just modernize, but it simplifies and accelerates. This article has an overview of the new features with code examples to boot! 📊 QuickSight QuickStart: Chris K. shows how to go from a blank AWS account to a fully deployed BI environment in one command. No click-heavy setup. Just dashboards, data, and fast insights. 🧠 RAG Meets Computational Linguistics: Here, I discuss computational linguistics concepts and how Retrieval-augmented generation becomes far more powerful when paired with knowledge of and clever implementation of the fundamentals. This isn’t just theory. It’s how you build smarter RAG systems. We're always aiming to ship ideas that matter and push us forward, as well as systems that scale. Dive into the details or reach out if you're solving similar challenges! https://lnkd.in/gTzHibvk https://lnkd.in/gawTbraM #Angular #AWSQuickSight #NLP #RAG #DataEngineering #SoftwareDevelopment #MachineLearning #AI #CloudComputing #NewMathData

16

2 Comments

Ever since the Agent Framework was announced, many have been curious (and a bit concerned) about its impact on the future of Semantic Kernel. Shawn Henry addresses most of those questions here: https://lnkd.in/d6F6-nyM

73

3 Comments

Wonderful article, "Making AI Less Thirsty: Uncovering and addressing the secret water footprint of AI models" by Shaolei Ren et al posted here: https://lnkd.in/eHxZZ3fn Recent sustainability efforts around the rapid growth in AI has been its carbon footprint, but an equally important aspect that is often overlooked is it's water footprint. Given the number of data centers currently growing or newly being built in areas of the world already under water stress, the sustainability community can't afford to ignore this impact as they build broader sustainability standards and frameworks for AI. Highlighted in the article are estimates of this rapidly growing threat, including this comment: "according to the recent U.S. datacenter energy report, the total annual onsite water consumption by U.S. datacenters in 2028 could double or even quadruple the 2023 level, reaching approximately 150–280 billion liters and further stressing the water infrastructures."  Well worth the deep dive into the calculations to understand the details around how thirsty AI is. https://lnkd.in/eJFnuSQh #SustainableWater #SustainableArchitectures #SustainableIT #ResponsibleAI #GreenLeadership #CIO #CTO #Innovation #ClimateAction

14

2 Comments