Large language models excel at many tasks but often incur high inference costs during deployment. To mitigate hallucination, many systems use a knowledge graph to enhance retrieval-augmented generation (KG-RAG). However, the large amount of retrieved knowledge contexts increase these inference costs further. A promising solution to balance performance and cost is LLM routing, which directs simple queries to smaller LLMs and complex ones to larger LLMs. However, no dedicated routing methods currently exist for RAG, and existing training-based routers face challenges scaling to this domain due to the need for extensive training data. We observe that the score distributions produced by the retrieval scorer strongly correlate with query difficulty. Based on this, we propose an extremely simple yet effective routing framework, the first specifically designed for KG-RAG that efficiently balances performance and cost in a plug-and-play manner. It delivers over 3x higher routing effectiveness while reducing runtime to less than 0.001x compared to existing methods. Our code is available at https://github.com/hrwang00/SkewRoute.

Although Large Language Models achieve strong success in many tasks, they still suffer from hallucinations and knowledge deficiencies in real-world applications. Many knowledge graph-based retrieval-augmented generation (KG-RAG) methods enhance the quality and credibility of LLMs by leveraging structure and semantic information in KGs as external knowledge bases. However, these methods struggle to effectively incorporate structure information, either incurring high computational costs or underutilizing available knowledge. Inspired by smoothing operations in graph representation learning, we propose path pooling, a simple, train-free strategy that introduces structure information through a novel path-centric pooling operation. It seamlessly integrates into existing KG-RAG methods in a plug-and-play manner, enabling richer structure information utilization. Extensive experiments demonstrate that incorporating the path pooling into the state-of-the-art KG-RAG method consistently improves performance across various settings while introducing negligible additional cost. Code is coming soon at https://github.com/hrwang00/path-pooling.

Large Language Models (LLMs) demonstrate strong reasoning abilities but face limitations such as hallucinations and outdated knowledge. Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) addresses these issues by grounding LLM outputs in structured external knowledge from KGs. However, current KG-based RAG frameworks still struggle to optimize the trade-off between retrieval effectiveness and efficiency in identifying a suitable amount of relevant graph information for the LLM to digest. We introduce SubgraphRAG, extending the KG-based RAG framework that retrieves subgraphs and leverages LLMs for reasoning and answer prediction. Our approach innovatively integrates a lightweight multilayer perceptron with a parallel triple-scoring mechanism for efficient and flexible subgraph retrieval while encoding directional structural distances to enhance retrieval effectiveness. The size of retrieved subgraphs can be flexibly adjusted to match the query's need and the downstream LLM's capabilities. This design strikes a balance between model complexity and reasoning power, enabling scalable and generalizable retrieval processes. Notably, based on our retrieved subgraphs, smaller LLMs like Llama3.1-8B-Instruct deliver competitive results with explainable reasoning, while larger models like GPT-4o achieve state-of-the-art accuracy compared with previous baselines -- all without fine-tuning. Extensive evaluations on the WebQSP and CWQ benchmarks highlight SubgraphRAG's strengths in efficiency, accuracy, and reliability by reducing hallucinations and improving response grounding.

LAS VEGAS – Most baseball fans won't forget the controversial call in Game 6 of the 2019 World Series between the Houston Astros and the Washington Nationals when runner Trea Turner was ruled out because of interference. Umpire accuracy is a frustration for fans and players in nearly every game. This season, MLB has launched so-called "robot umpires" in 11 Pacific Coast League Triple-A teams, putting it one step away from reaching the major leagues, to improve accuracy and reduce delays. The automated balls and strikes system (ABS) debuted in a Las Vegas Aviators' game earlier this month. As cool and bizarre as it would be to see "Jetsons"-style robots on the field, most fans won't notice the actual device -- eight surveillance-looking cameras at the top of the bleachers.

We all have our own tried and true ways to boost productivity and focus. I grew up in a house filled with 60s music, video games, and Cubs baseball broadcasts. Perhaps it's not surprising, in a career of hopping around between computational physics, application development, data science and engineering but always coding, that I have come to love working with the game on in the background -- especially during those Friday afternoon games at Wrigley. To me, these things just go together. The same goes for quantum computing and machine learning (ML).

Snell, second from left, comes out of the game against the Dodgers in the 6th inning in Game 6 of the World Series at Globe Life Field on October 27, 2020 in Arlington, Texas. I'm a lifelong Dodgers fan and I waited for 32 years for the team to win another World Series. But during this period of time, the sport has certainly seen much change. With the availability of huge amounts of data, sophisticated computers and advanced analytics, the strategies have become increasingly based on the numbers. It seems that AI (Artificial Intelligence) has dominated the decision making process.

ARLINGTON, TEXAS - OCTOBER 27:Rays pitcher Blake Snell, second from left, comes out of the game ... [ ] against the Dodgers in the 6th inning in Game 6 of the World Series at Globe Life Field on October 27, 2020 in Arlington, Texas. I'm a lifelong Dodgers fan and I waited for 32 years for the team to win another World Series. But during this period of time, the sport has certainly seen much change. With the availability of huge amounts of data, sophisticated computers and advanced analytics, the strategies have become increasingly based on the numbers. It seems that AI (Artificial Intelligence) has dominated the decision making process.

Major League Baseball is entering uncharted waters with the start of its COVID-abridged season today. Nobody's really sure if the 60-game season will even be able to get through the World Series without disruption by the pandemic's spread. However, one crowd-sourced AI system already has a pretty good guess as to who will be taking home the Commissioner's Trophy. The folks at Unanimous AI have been making high profile predictions like these since 2016, when their UNU platform correctly figured 11 of 15 winners for that year's Academy Awards. In 2017, the company followed up by correctly guessing the Kentucky Derby's top four finishers -- in order, no less -- and in 2019, correctly figured that the Houston Astros would make it to the series (though nobody could have seen the Nat's miraculous postseason run coming). "The fundamental core of our system is a technology that captures input from groups of people by connecting them together in real time using AI algorithms modeled after swarms," Dr. Louis Rosenberg, Unanimous' founder and chief scientist, told Engadget.

One thing I get asked about a lot is weighting. What do I need to worry about? By popular demand, I recently put together a lunch-and-learn at my company to help address all these questions. The goal was to be applicable to a large audience, (e.g., with a gentle introduction), but also some good technical advice/details to help practitioners. This blog was adapted from that presentation. Before we talk about weighting, we should all get on the same page about what a model is, what they are used for, and some of the common issues that modelers run into.

I am roaming the ravaged streets of Los Angeles while Patrick Mahomes is in my ears; directing me to watch my back. Despite the best efforts of the Kansas City Chiefs quarterback and the reigning NFL MVP, I am no match for the sniper sitting atop a nearby roof who takes me out in front of a police car. "I told you to watch out," Mahomes said. Thankfully I have a next time because Mahomes and I are playing "Call of Duty: Black Ops 4 'Operation Grand Heist,'" which was released this week. Mahomes was in Los Angeles to get a tour of Treyarch, a video game developer in Santa Monica, which developed the game and six other "Call of Duty" titles dating to "Call of Duty 2: Big Red One" in 2005.