/
划分小节”,那小节标题要从原文里找吗?不对,原文里没有现成的小节标题,那怎么办?哦,不对,仔细看原文的内容,原文的段落其实是连贯的,但可以按照内容逻辑划分小节,但标题必须用原文里的语句吗?不对,不能增删,所以是不是应该先把原文的内容拆分,然后用h2/h3来包裹对应的核心句作为小节标题? 等下,再仔细看原文的内容: 原文第一句是问题:“Can ranking a batch of antipodal query pairs recover a gradient direction with fewer evaluations than using within-pair signs alone?” 然后是研究内容:“We study fixed-center recovery in $d$ dimensions for a linear signal plus an unknown differentiable even nuisance, counting function evaluations and feedback rounds separately.” 接下来:“Shared pair offsets create an information hierarchy: Mirror retains within-pair signs, Reverse couples reversed cross-pair orders to cancel offsets, and Full enforces all ranking inequalities. Relative shell variation is the nuisance range over the query sphere normalized by the linear-signal scale.” 然后是Reverse的结果:“When it is at most $c\varepsilon/\sqrt{d}$, Reverse localizes every consistent direction to angular accuracy $\varepsilon$ using $\widetilde{O}(d+\sqrt{d/\varepsilon})$ evaluations in one round at fixed confidence, without a variation bound supplied to the decoder. This matches a lower bound up to logarithmic factors, versus the $Θ(d/\varepsilon)$ one-round sign-only minimax cost on the same class.” 然后是正定二次型的情况:“Over well-conditioned positive-definite quadratics with no common relative-curvature bound, the $R$-round minimax cost is instead $Θ(dR(\varepsilon^{-1/R}-1))$, even with cumulative rankings.” 然后是扩展和实验:“We extend validity beyond exact evenness under supplied error bounds. Matched-endpoint experiments certify information beyond signs and short cycles. On two ridge tasks under joint evaluation and round caps, short-cycle recovery attains lower terminal mean gaps than the tested batch-capable controls. An additional optimization benefit from fitting the complete ranking remains unestablished.” 哦,不对,不能添加内容,所以小节标题不能用原文没有的,那是不是可以把这些核心的句子作为h2/h3?比如,比如第一个h2可以是研究问题?不对,原文没有,那会不会原文其实不需要分太多h2?或者,有没有可能原文是一段,但需要拆分p?不对,规则2说每个自然段落用p包裹,长文用h2/h3划分小节。那首先,自然段落怎么划分?原文是一整段,但可以按照内容逻辑拆分自然段落?不对,自然段落是原文里的段落,原文给的是一整段,那是不是需要按照语义拆分p? 等下,再理清楚:首先,不能增删改内容,所以所有文字都要保留,只是加HTML标签。那首先: 首先,开头的问题,可以作为第一个部分?比如,先有一个p包含问题和研究设置?然后,接下来是信息层级和相对壳层变化的定义,作为一个部分?然后是Reverse的结果和对比,作为一个部分?然后是正定二次型的下界,作为部分?然后是扩展、实验和结论,作为部分? 不对,那h2的标题怎么办?哦,哦,对了,是不是可以把原文里的每个核心论点的开头作为h2?比如: 第一个h2:“Research Question”?不对,不行,原文是英文,不能加中文,也不能加原文没有的词。哦!哦对,原文是英文,所以所有标签里的内容都必须是原文里的,不能自己加标题。那是不是原文其实不需要h2/h3?不对,规则说长文用h2/h3划分小节,那如果内容长的话,需要划分,那怎么划分?哦,等下,看原文里的内容,有“Shared pair offsets create an information hierarchy”这句话,可以作为h2?然后“When it is at most $c\varepsilon/\sqrt{d}$”可以作为h2?不对,或者把关键的句子作为h3? 等下,再仔细看要求:“保留原文所有信息,不得增删改内容”,所以添加的标题如果原文没有的话,就是增内容了,对吧?哦,原来如此!那是不是h2/h3的标题必须是原文中的句子,不能自己添加?对,这样才符合不增删的要求。那这样的话,我们可以把原文中的核心主题句作为h2,然后相关的内容作为p或者列表等。 那现在开始排版: 首先,原文的第一句是问题,然后是研究的对象,这部分可以作为第一个p?或者把问题作为开头,然后:
Can ranking a batch of antipodal query pairs recover a gradient direction with fewer evaluations than using within-pair signs alone? We study fixed-center recovery in $d$ dimensions for a linear signal plus an unknown differentiable even nuisance, counting function evaluations and feedback rounds separately.
然后,接下来的内容是关于信息层级和相对壳层变化的,这里有个并列的内容,就是信息层级的三个类型,可以用ul?哦,原文里是“Shared pair offsets create an information hierarchy: Mirror retains within-pair signs, Reverse couples reversed cross-pair orders to cancel offsets, and Full enforces all ranking inequalities.” 这里可以把这个句子作为h2,然后把三个类型作为ul?不对,原文里的冒号后面是三个内容,所以可以这样:Shared pair offsets create an information hierarchy
- Mirror retains within-pair signs
- Reverse couples reversed cross-pair orders to cancel offsets
- Full enforces all ranking inequalities
Relative shell variation is the nuisance range over the query sphere normalized by the linear-signal scale.
对,这样是对的,因为原文的冒号后面就是三个并列的,所以转换成ul列表,没有增删内容。 然后接下来的内容是关于Reverse的结果,这里可以把这个部分的开头核心句作为h2?比如“When it is at most $c\varepsilon/\sqrt{d}$”不对,应该是完整的句子,哦,原文里“When it is at most $c\varepsilon/\sqrt{d}$, Reverse localizes every consistent direction to angular accuracy $\varepsilon$ using $\widetilde{O}(d+\sqrt{d/\varepsilon})$ evaluations in one round at fixed confidence, without a variation bound supplied to the decoder.” 这部分可以作为p?或者把前面的结果和对比作为一个部分:When it is at most $c\varepsilon/\sqrt{d}$, Reverse localizes every consistent direction to angular accuracy $\varepsilon$ using $\widetilde{O}(d+\sqrt{d/\varepsilon})$ evaluations in one round at fixed confidence, without a variation bound supplied to the decoder. This matches a lower bound up to logarithmic factors, versus the $Θ(d/\varepsilon)$ one-round sign-only minimax cost on the same class.
然后接下来的关于正定二次型的内容:“Over well-conditioned positive-definite quadratics with no common relative-curvature bound, the $R$-round minimax cost is instead $Θ(dR(\varepsilon^{-1/R}-1))$, even with cumulative rankings.” 这个可以作为p,或者把核心作为h2?不对,不需要,因为可以作为p。 然后最后一部分:“We extend validity beyond exact evenness under supplied error bounds. Matched-endpoint experiments certify information beyond signs and short cycles. On two ridge tasks under joint evaluation and round caps, short-cycle recovery attains lower terminal mean gaps than the tested batch-