🔬
ACL SRW、ICML 工作坊還是 AACL？
研究者討論將 ARR 分數為 3、2.5、2.5（元評分 2.5）的修訂 ML 論文提交至 ACL SRW、ICML 工作坊，或等待 AACL。擔憂領域發展迅速、錄取機會低，以及工作坊聲望與及時發表之權衡。教授建議謹慎考慮工作坊。
Search
Tag: #nlp54 results
🔬
ACL 2026 視覺語言模型機制解釋賽道選擇
視覺語言模型機制解釋性論文（注意力頭、logit lens、因果干預）瞄準 ACL 2026。混淆特殊「NLP 模型可解釋性」主題賽道與標準「模型解釋與分析」賽道。尋求差異與競爭力建議。
AI Writing's Semantic Ablation Flaw
An opinion piece highlights 'semantic ablation' as a subtractive bias making AI writing generic and boring. It contrasts this with well-known 'hallucinations' which are additive errors. The issue poses dangers beyond mere blandness.
FoSS: GFlowNets for Dynamic Span LMs
FoSS introduces a GFlowNets framework for generating text via dynamic span vocabularies in a DAG-structured state space. It enables flexible segmentation of retrieved text and explores diverse compositional paths. Empirically, it boosts MAUVE scores by 12.5% and excels in knowledge tasks.