❌

Reading view

Evaluating AI in leukocyte classification: performance of the AI system against 15 morphology experts

npj Digital Medicine, Published online: 11 April 2026; doi:10.1038/s41746-026-02601-w

Evaluating AI in leukocyte classification: performance of the AI system against 15 morphology experts
  •  

Genetic mutation and dysfunction of AT2 cells drive B(a)P/LPS-induced inflammation-related lung tumorigenesis: evidence and mechanism of autophagy

Acta Biochim Biophys Sin (Shanghai). 2026 Mar 25. doi: 10.3724/abbs.2025238. Online ahead of print.

ABSTRACT

The environmental pollutant benzo(a)pyrene (B(a)P), a representative polycyclic aromatic hydrocarbon (PAH), is a recognized carcinogen, and chronic pulmonary inflammation is closely associated with lung carcinogenesis. Although alveolar type 2 (AT2) cells are the origin of lung adenocarcinoma, the genetic and functional changes in AT2 cells and the mechanisms involved in inflammation-related lung tumorigenesis have not been elucidated. Here, C57BL/6J mice are exposed to B(a)P and the inflammatory irritant lipopolysaccharide (LPS) to establish a model of inflammation-related lung tumorigenesis. Single-cell RNA sequencing is performed on lung tissues. DNA mutations in AT2 cells are analyzed via whole-exome sequencing. The protein expression of AT2 cells in lung cancer tissue is determined by immunofluorescence staining. The results reveal that LPS promotes B(a)P-induced lung tumorigenesis; in the whole lungs of B(a)P/LPS, a decreased proportion, altered differentiation trajectory, and increased gene mutation number in AT2 cells are observed. Additionally, in B(a)P/LPS-treated lung cancer tissue, the levels of Ξ³-H2AX DNA damage and the proliferation marker Ki67 in AT2 cells are increased, whereas the levels of differentiation markers are decreased. Single-cell RNA transcriptomics reveals that the autophagy-related genes Foxo3 and Ppp2r5, which are enriched in the PI3K-Akt pathway, and the autophagy-related genes in AT2 cells in lung cancer are decreased in the B(a)P/LPS group. Thus, chronic inflammation promotes DNA damage, gene mutation and dysfunction in AT2 cells, and decreased autophagy in AT2 cells may be an important mechanism for inflammation-related lung tumorigenesis.

PMID:41952558 | DOI:10.3724/abbs.2025238

  •  

100x Cost & Latency Reduction: Performance Analysis of AI Query Approximation using Lightweight Proxy Models

arXiv:2603.15970v5 Announce Type: replace-cross Abstract: Several data warehouse and database providers have recently introduced extensions to SQL called AI Queries, enabling users to specify functions and conditions in SQL that are evaluated by LLMs, thereby broadening significantly the kinds of queries one can express over the combination of structured and unstructured data. LLMs offer remarkable semantic reasoning capabilities, making them an essential tool for complex and nuanced queries that blend structured and unstructured data. While extremely powerful, these AI queries can become prohibitively costly when invoked thousands of times. This paper provides an extensive evaluation of a recent AI query approximation approach that enables low cost analytics and database applications to benefit from AI queries. The approach delivers >100x cost and latency reduction for the semantic filter operator and also important gains for semantic ranking. The cost and performance gains come from utilizing cheap and accurate proxy models over embedding vectors. We show that despite the massive gains in latency and cost, these proxy models preserve accuracy and occasionally improve accuracy across various benchmark datasets, including the extended Amazon reviews benchmark that has 10M rows. We present an OLAP-friendly architecture within Google BigQuery for this approach for purely online (ad hoc) queries, and a low-latency HTAP database-friendly architecture in AlloyDB that could further improve the latency by moving the proxy model training offline. We present techniques that accelerate the proxy model training.
  •  

100x Cost & Latency Reduction: Performance Analysis of AI Query Approximation using Lightweight Proxy Models

arXiv:2603.15970v4 Announce Type: replace-cross Abstract: Several data warehouse and database providers have recently introduced extensions to SQL called AI Queries, enabling users to specify functions and conditions in SQL that are evaluated by LLMs, thereby broadening significantly the kinds of queries one can express over the combination of structured and unstructured data. LLMs offer remarkable semantic reasoning capabilities, making them an essential tool for complex and nuanced queries that blend structured and unstructured data. While extremely powerful, these AI queries can become prohibitively costly when invoked thousands of times. This paper provides an extensive evaluation of a recent AI query approximation approach that enables low cost analytics and database applications to benefit from AI queries. The approach delivers >100x cost and latency reduction for the semantic filter operator and also important gains for semantic ranking. The cost and performance gains come from utilizing cheap and accurate proxy models over embedding vectors. We show that despite the massive gains in latency and cost, these proxy models preserve accuracy and occasionally improve accuracy across various benchmark datasets, including the extended Amazon reviews benchmark that has 10M rows. We present an OLAP-friendly architecture within Google BigQuery for this approach for purely online (ad hoc) queries, and a low-latency HTAP database-friendly architecture in AlloyDB that could further improve the latency by moving the proxy model training offline. We present techniques that accelerate the proxy model training.
  •  

Thinking in Streaming Video

arXiv:2603.12938v1 Announce Type: cross Abstract: Real-time understanding of continuous video streams is essential for interactive assistants and multimodal agents operating in dynamic environments. However, most existing video reasoning approaches follow a batch paradigm that defers reasoning until the full video context is observed, resulting in high latency and growing computational cost that are incompatible with streaming scenarios. In this paper, we introduce ThinkStream, a framework for streaming video reasoning based on a Watch--Think--Speak paradigm that enables models to incrementally update their understanding as new video observations arrive. At each step, the model performs a short reasoning update and decides whether sufficient evidence has accumulated to produce a response. To support long-horizon streaming, we propose Reasoning-Compressed Streaming Memory (RCSM), which treats intermediate reasoning traces as compact semantic memory that replaces outdated visual tokens while preserving essential context. We further train the model using a Streaming Reinforcement Learning with Verifiable Rewards scheme that aligns incremental reasoning and response timing with the requirements of streaming interaction. Experiments on multiple streaming video benchmarks show that ThinkStream significantly outperforms existing online video models while maintaining low latency and memory usage. Code, models and data will be released at https://github.com/johncaged/ThinkStream
  •  
❌