bioengineer.org /1 month ago

Hypergraph Adapters Push Parameter-Efficient Multimodal Fine-Tuning to New Heights

Researchers have introduced HGA-Net, a hypergraph-based adapter that connects related examples within each mini-batch and achieves near-perfect fine-grained recognition with only 1...

Read source
bioengineer.org /1 month ago

Graph Networks Learn to Read the Hidden Tactics Inside Sports Videos

A new spatio-temporal graph network models players and the ball as a dynamic graph, detecting key match events and tactical patterns with record accuracy and real-time speed.

Read source
towardsdatascience.com /1 month ago

GraphRAG with TypeSafe Jev: A System One Approach to Scalable Knowledge Graphs

How calibrated decision models can handle high-frequency graph decisions while LLMs remain focused on reasoning, synthesis, and open-ended generation. The post GraphRAG with TypeSa...

Read source
huxiangqian.medium.com /1 month ago

Can Neural Networks Learn Directly on Low-Bit Weights?

Revisiting gradient-directed Monte Carlo — from molecular design to a 16-value weight grid.Continue reading on Medium »

Read source
machinelearningmastery.com /1 month ago

Evaluating Graph-RAG vs. Standard RAG: A Hallucination Benchmark on Fact-Dense Queries

In this article, you will learn how to benchmark a deterministic 3-Tiered Graph-RAG system against a standard vector RAG pipeline on fact-dense queries, and what...

Read source
pandaily.com /3 weeks ago

Hi3D Open-Sources Hi3D v1.0, a Sparc3D-Based Model That Refines 3D Meshes Up to 1536 Resolution

Hi3D has released inference code and weights for Hi3D v1.0, which turns a coarse mesh and a reference image into detailed geometry at a sparse grid resolution of up to 1536 cubed.

Read source
towardsdatascience.com /3 weeks ago

Building and Visualising Kolmogorov–Arnold Networks

Exploring what happens when neural networks learn functions instead of weights. The post Building and Visualising Kolmogorov–Arnold Networks appeared first on Towards Data Science.

Read source
medium.com /1 month ago

Build a Dynamic Parallel Multi-Agent System with LangGraph and Hugging Face

Let the AI decide which agents to run — and run the right ones simultaneouslyContinue reading on Data Science Collective »

Read source
journals.plos.org /1 month ago

MEGH-DDI: MolT5-driven edge-aware graph learning with Hop2Token representation for drug–drug interaction prediction

by Li Zhou, Yun Zuo, Wenying He, Zhaohong Deng Drug-drug interactions (DDIs) are critical in pharmacology, as synergistic or antagonistic effects between co-administered drugs can...

Read source
techmeme.com /2 weeks ago

Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B...

Rikiya Takehi: Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B's weights to make it better at math  —  TL;D...

Read source
kdnuggets.com /1 week ago

Building Graph-Enhanced GenAI Applications with graphrag-toolkit

Learn how graph-enhanced RAG connects related facts that plain vector search misses, then build a multi-hop research assistant with AWS Labs' open-source graphrag-toolkit, Neo4j, a...

Read source
marktechpost.com /2 weeks ago

Zyphra AI Introduces Zyphra Hierarchical Attention (ZHA): A Path to True Subquadratic Attention

Zyphra Hierarchical Attention (ZHA) is a tree-indexed sparse attention method that retrofits frozen LLMs with 4B tokens of distillation, matching dense attention on reasoning and 2...

Read source
bioengineer.org /1 month ago

Diffusion Meets Graphs: New AI Model Predicts Spatio-Temporal Data With Built-In Uncertainty

A new conditioned diffusion-based graph attention network from the University of Naples Parthenope delivers probabilistic spatio-temporal forecasts with calibrated uncertainty, ran...

Read source
bioengineer.org /1 month ago

Quantum Neural Networks Get a Rework That Makes Them Easier to Train and Build

Researchers in Vienna have redesigned dissipative quantum neural networks so each building block can implement any quantum channel, and shown that the easily measured Hilbert-Schmi...

Read source
pandaily.com /2 weeks ago

HongYun's HyperLM V2.5 Adds Native Support for Huawei Ascend 950DT and Atlas 850E SuperPoD

HongYun's HyperLM V2.5 adapts to Huawei's Ascend 950DT and Atlas 850E, adding UnifiedBus 2.0 support, mxFP4 tuning, vNPU slicing and prefill-decode split inference.

Read source
rna-seqblog.com /1 month ago

GP-DHT uses single-cell RNA sequencing to map gene regulatory networks

Single-cell RNA sequencing combined with GP-DHT helps predict transcription factor and target gene relationships, providing a...

Read source
journals.plos.org /1 month ago

A graph-attentive GAN for rare-cell-aware single-cell RNA-seq data generation

by Ritwik Ganguly, Sana Aafrine, Sk Md Mosaddek Hossain, Sumanta Ray A central challenge in downstream single-cell RNA sequencing (scRNA-seq) analysis is the high-dimensional, sma...

Read source
iguides.ru /1 month ago

Российские исследователи научили графовые нейросети работать с миллиардами связей

AI VK Research разработала метод, который позволяет масштабировать графовые нейросети на сотни миллионов объектов и миллиарды связей.

Read source
towardsdatascience.com /1 week ago

A Practitioner's Guide to 6 Agentic GraphRAG Patterns

How autonomous agents turn static knowledge graphs into living Context Graphs that remember decisions, user intent, and operational history The post A Practitioner's Guide to 6 Age...

Read source
pandaily.com /1 month ago

Xiaomi Details HySparse2 for MiMo-V3: Two-Level KV Sharing Cuts 1M-Token Prefill FLOPs About 5× vs Hybrid SWA

Xiaomi's LLM-Core team published HySparse2, a hybrid sparse-attention design with two-level KV sharing aimed at MiMo-V3-class agentic models, reporting about 5× lower 1M-token pref...

Read source
bioengineer.org /2 days ago

Compact Hirota Bilinear Neural Network Achieves Competitive Vietnamese Sentiment Prediction

A lightweight deep learning model incorporating Hirota bilinear interactions achieved competitive Vietnamese sentiment prediction performance using only 6,915 trainable parameters.

Read source
quantumcomputingreport.com /3 weeks ago

Gothenburg and Chalmers Researchers Develop Modular Graph Neural Decoder for Surface Code Memory and Lattice Surgery

Researchers from the University of Gothenburg and Chalmers University of Technology have developed a modular neural decoder for surface code quantum error correction (QEC). This de...

Read source
dzone.com /1 month ago

RAG Is Not Enough: The Rise of Enterprise Knowledge Graphs for AI Systems

Retrieval-augmented generation has become a standard pattern for grounding large language models in enterprise data. A typical implementation converts documents into embeddings, st...

Read source

Turn fresh research into a full content calendar

Use SocialBu to discover ideas, generate post drafts, and schedule them across your social channels.