OffNet Newsroom

Daily topic roundup

AI / ML

Monday, September 14, 2026 · 5 stories, curated & summarized — click any story for the source.

Share this roundup on LinkedIn Full edition
AWS What's New awsdatabase

AWS Entity Resolution introduces record-level ML confidence scores

AWS Entity Resolution now assigns individual confidence scores to each matched record rather than applying a uniform group-level score. This change allows systems to distinguish high-quality matches from borderline ones, enabling differentiated thresholding for downstream activation. Previously, a single threshold limited the volume of resolved identities that could be safely processed.

  • Use per-record scores to lower activation thresholds for high-confidence matches without increasing false positives
  • Retire uniform group-level confidence checks in favor of granular record-level filtering logic
  • Improve identity resolution throughput by activating borderline matches that previously failed strict group thresholds
TRADE-OFFEntity Resolution Scoring ShiftOld Group LevelUniform score for all recordsSingle threshold for activationLimits total identity volumeNew Record LevelIndividual ML confidence per recordGranular filtering logic per matchHigher throughput…vs
Read at source →
arXiv cs.AI researchai

KGE Edits Often Displace Correct Answers Despite Successful Target Promotion

Research shows that editing knowledge graph embeddings to promote a desired answer frequently displaces other correct answers from the result list. Standard locality tests based on parameter reuse fail to detect this ranking degradation. The study introduces a broader audit scope and identifies geometric conditions required to preserve existing scores during updates.

  • Promoting a target answer in KGE models often harms other correct answers.
  • Parameter-locality tests are insufficient for detecting ranking side-effects.
  • Only ~23% of edits on FB15k-237 preserve other correct answer ranks.
  • Geometric conditions exist to strictly preserve scores during updates.
BY THE NUMBERSKGE Edit Side Effects23%Edits preserving other correct ranksMost edits displace valid answers despite target promotion
Read at source →
Hugging Face Blog llmaiml ↺ since 09-10

IBM releases SOTA Granite Time Series PatchTST-FM-r2 on Hugging Face

IBM Research has released the PatchTST-FM-r2 model, currently achieving state-of-the-art performance on time series tasks. The model is available via Hugging Face under a commercial-friendly license, allowing for broad enterprise adoption. This release targets practitioners needing high-accuracy forecasting capabilities without restrictive licensing barriers.

  • IBM's PatchTST-FM-r2 is the new SOTA for time series forecasting.
  • Commercial-friendly license enables unrestricted enterprise usage.
  • Model is immediately accessible via Hugging Face Hub.
  • Targeted at high-accuracy needs in operational tech stacks.
Read at source →
InfoQ generaldevops ↺ since 09-12

NVIDIA PAIR Beta Distributes Local AI Inference Across Multi-GPU Networks

NVIDIA has released a beta version of the Personal AI Router (PAIR), a tool designed to aggregate inference capacity from multiple local computers. It automatically routes AI requests across available hardware on a local network, specifically targeting multi-agent workloads. This approach helps prevent single GPU bottlenecks by distributing independent model calls across the available local compute resources.

  • PAIR beta enables horizontal scaling of local AI inference across multiple machines.
  • Designed to offload multi-agent workloads that overwhelm single GPU setups.
  • Automatically distributes requests without manual load balancing configuration.
  • Leverages existing local network hardware to boost aggregate inference throughput.
Read at source →
The Register general ↺ since 09-13

DOJ eyes Nvidia-Groq acquihire, but market alternatives already absorb talent

The Department of Justice is scrutinizing Nvidia's $20 billion acquisition of Groq, though the deal is considered effectively irreversible. Even if regulators attempted to unwind the transaction, the AI hardware landscape has shifted to accommodate Groq's workforce without a merger. A growing list of competitors is prepared to absorb Groq's engineers and technology independently.

  • DOJ is monitoring the $20B Nvidia-Groq deal despite its advanced status
  • Unwinding the acquisition is viewed as practically impossible by analysts
  • Competitors are ready to absorb Groq's talent without a merger
  • Market dynamics already neutralize potential anti-competitive effects
Read at source →
↑