paper_id	claim	claim_status	evidence_status	source_depth	source_quote	page_or_section	taxonomy_fit	audit_status
2405.13407v1	Researchers have begun exploring adaptive or conditional residuals as a means to improve the representational power of models.	supported	has evidence row	full-text	Using the Huggingface Transformers library (Apache License 2.0), we enhance the BERT [ 8] baseline model by integrating these components, utilizing the bert-base-uncased variant.	p3	in-scope: taxonomy category match	pass; full-text verified; report=audit_report.md
2509.14199v2	•Gated Residual Tokenization (GRT):We present a two-stage framework for accelerating and reducing tokenization in dense video settings: 1.Motion-Compensated Gated Inter-Tokenization filters out uninformative patches before tokenization using per-pixel motion masks.	supported	has evidence row	full-text	Our 0.5B-parameter model achieves an MOS of 2.50, outperforming all baselines—including the larger 7B-parameter LLaV A-Video (1.47) and both 0.5B and 7B variants of LLaV A-OV and LLaV A-SI.	p3	in-scope: taxonomy category match	pass; full-text verified; report=audit_report.md
2008.11865v1	We shall demonstrate empirically that these matrices cause various spectral features: 1.In effect, we are introducing into deepnets constructs familiar in Multivariate Analysis of Variance (MANOVA), where the class/cross-class index structure would be called a two-way categorical layout.	supported	has evidence row	full-text	Moreover, we will distinguish between vectorsvi;c;c0, wherec=c0andc6=c0.1 1.11 Cause attribution As the introduction has shown, various spectral features have been observed in the literature.	p7	in-scope: taxonomy category match	pass; full-text verified; report=audit_report.md
2504.13990v1	To summarize, the main contributions of this study are as follows: 1) We propose a PC-DeepNet framework using the PI-DNN model to handle the variation in the number and order of satellite measurements and minimize the positioning error.	supported	has evidence row	full-text	They claim an improvement of position accuracy from 81.3m to 23.3m compared to the conventional method [32] which does not satisfy user requirement.	p4	in-scope: taxonomy category match	pass; full-text verified; report=audit_report.md
2409.15161v2	F RAMEWORK In this paper, we introduce a new framework called “KAMoE” Figure 1, based on Gated Residual KolmogorovArnold Networks (GRKAN) introduced in our previous work [22].	supported	has evidence row	full-text	Through extensive experiments on digital asset markets and real estate valuation, we demonstrate that KAMoE consistently outperforms traditional MoE architectures across various tasks and model types.	p2	in-scope: taxonomy category match	pass; full-text verified; report=audit_report.md