Understanding the Energy Scaling of Large Language Model Inference Across Context Lengths and Attention Architectures
本文通过研究不同注意力机制和上下文长度下的能量消耗，解决了大型语言模型推理过程中的能耗问题，提供了选择节能架构的指导。
0 citationsRead paper
Selected work
Representative Papers
Property Graph Techniques in Relational Databases
This study addresses the challenge of unifying attribute graph models and SQL querying within relational databases. The authors propose reinterpreting SQL foreign key semantics as reference keys, enabling natural modeling of labeled property graphs directly on standard relational table structures. They further extend SQL to support efficient graph data insertion and complex pattern matching. This approach achieves deep integration of relational and graph models within a single system without requiring an additional storage engine. Experimental results demonstrate that the proposed method effectively enables graph structure construction and advanced graph querying capabilities, significantly enhancing relational databases’ support for graph-oriented operations.
0 citationsRead paper
From Relational and Property Graph Data to Large Language Models
This work addresses the need for unified and efficient knowledge provisioning in large language models by proposing a novel architecture that integrates relational and property graph data models. The approach leverages record addresses from log files as immutable reference values in place of traditional foreign keys, enabling efficient graph-style link traversal instead of costly join queries while natively supporting triple-based knowledge representation. The resulting unified knowledge service framework combines the structural rigor of relational models with the flexible associative capabilities of graph models, significantly enhancing knowledge retrieval efficiency and effectively supporting knowledge integration and invocation in generative AI systems.
0 citationsRead paper
WorkstreamBench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance
Existing benchmarks inadequately assess the ability of large language model (LLM) agents to generate complete spreadsheets end-to-end in financial contexts—such as financial modeling or scenario analysis—being largely confined to question answering or single-formula editing. This work proposes the first end-to-end spreadsheet generation evaluation framework tailored to real-world financial workflows, evaluating performance along three core dimensions: accuracy, formula correctness, and formatting compliance, while also introducing professional criteria such as readability and modifiability for the first time. The framework integrates both human and automated evaluation methods for comprehensive assessment. Experimental results demonstrate that Claude-family models achieve the strongest performance among current LLMs, yet still fall significantly short of expert human practitioners on complex tasks, revealing critical limitations of contemporary LLM agents in authentic financial settings.
0 citationsRead paper
Spectral Convolution on Orbifolds for Geometric Deep Learning
This work addresses the challenge of performing effective geometric deep learning on non-Euclidean data endowed with orbifold structures—quotient spaces that may contain singularities due to symmetry operations. Recognizing that existing methods struggle to handle such topological complexities, the authors propose the first extension of spectral convolution to orbifolds, thereby establishing foundational building blocks for geometric deep learning on this class of spaces. This advancement broadens the range of topological structures amenable to geometric deep learning and offers a novel framework for modeling data exhibiting both symmetries and singularities. The efficacy of the proposed approach is demonstrated through a case study in music theory, where it successfully captures the intrinsic geometry of real-world non-Euclidean data, highlighting its expressive power and practical potential.
0 citationsRead paper
Recent publications
Latest Papers
Understanding the Energy Scaling of Large Language Model Inference Across Context Lengths and Attention Architectures
本文通过研究不同注意力机制和上下文长度下的能量消耗，解决了大型语言模型推理过程中的能耗问题，提供了选择节能架构的指导。
0 citationsRead paper
Property Graph Techniques in Relational Databases
This study addresses the challenge of unifying attribute graph models and SQL querying within relational databases. The authors propose reinterpreting SQL foreign key semantics as reference keys, enabling natural modeling of labeled property graphs directly on standard relational table structures. They further extend SQL to support efficient graph data insertion and complex pattern matching. This approach achieves deep integration of relational and graph models within a single system without requiring an additional storage engine. Experimental results demonstrate that the proposed method effectively enables graph structure construction and advanced graph querying capabilities, significantly enhancing relational databases’ support for graph-oriented operations.
0 citationsRead paper
From Relational and Property Graph Data to Large Language Models
This work addresses the need for unified and efficient knowledge provisioning in large language models by proposing a novel architecture that integrates relational and property graph data models. The approach leverages record addresses from log files as immutable reference values in place of traditional foreign keys, enabling efficient graph-style link traversal instead of costly join queries while natively supporting triple-based knowledge representation. The resulting unified knowledge service framework combines the structural rigor of relational models with the flexible associative capabilities of graph models, significantly enhancing knowledge retrieval efficiency and effectively supporting knowledge integration and invocation in generative AI systems.
0 citationsRead paper
WorkstreamBench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance
Existing benchmarks inadequately assess the ability of large language model (LLM) agents to generate complete spreadsheets end-to-end in financial contexts—such as financial modeling or scenario analysis—being largely confined to question answering or single-formula editing. This work proposes the first end-to-end spreadsheet generation evaluation framework tailored to real-world financial workflows, evaluating performance along three core dimensions: accuracy, formula correctness, and formatting compliance, while also introducing professional criteria such as readability and modifiability for the first time. The framework integrates both human and automated evaluation methods for comprehensive assessment. Experimental results demonstrate that Claude-family models achieve the strongest performance among current LLMs, yet still fall significantly short of expert human practitioners on complex tasks, revealing critical limitations of contemporary LLM agents in authentic financial settings.
0 citationsRead paper
Spectral Convolution on Orbifolds for Geometric Deep Learning
This work addresses the challenge of performing effective geometric deep learning on non-Euclidean data endowed with orbifold structures—quotient spaces that may contain singularities due to symmetry operations. Recognizing that existing methods struggle to handle such topological complexities, the authors propose the first extension of spectral convolution to orbifolds, thereby establishing foundational building blocks for geometric deep learning on this class of spaces. This advancement broadens the range of topological structures amenable to geometric deep learning and offers a novel framework for modeling data exhibiting both symmetries and singularities. The efficacy of the proposed approach is demonstrated through a case study in music theory, where it successfully captures the intrinsic geometry of real-world non-Euclidean data, highlighting its expressive power and practical potential.
0 citationsRead paper