首页 > AI前沿 > GALA: Gallery-Level Alignment for Real-Life Vision Decoding from EEG

GALA: Gallery-Level Alignment for Real-Life Vision Decoding from EEG

arXiv机器学习 2026-11-18 18:40 21 阅读 查看原文
用户现在需要把给定的新闻正文排版成语义化的HTML片段,遵循那些规则,只输出HTML,不能有多余内容。首先,开头是一段,直接用

包裹。然后有没有需要分小节的?看内容,都是连贯的,不过先理清楚: 第一段是讲现有方法的问题,直接

。然后接下来讲提出的新设置,然后讲方法GALA,然后是实验结果。不对,先看每个自然段落,原文是一个大段落?不对,原文的正文是连续的,不过需要按内容划分吗?不对,规则说每个自然段落用

,长文用

/

划分小节。首先,原文的内容: 首先第一段是现状和问题,然后讲提出新设置,然后讲实验协议,然后讲方法GALA,然后讲实验结果。那可以划分小节吗? 首先,开头的现状部分?不对,先准确保留内容,不能改。首先: 第一个自然段(现状与问题)用

,然后可以加

提出的新设置

?不对,原文的表述是连贯的,等下,原文的内容: 1. 第一句到“limits practical device deployment.”是现状问题,用

。 2. 然后“In a practical Brain-Computer Interface...”到“than the zero-shot and batch evaluations in current use.”是提出新的设置,这里可以用

提出的新设置

?不对,规则说长文用

/

划分小节,然后

包裹段落。然后接下来“We evaluate it with a protocol...”是实验协议,然后“In this setting, we introduce GALA...”是方法介绍,然后“On THINGS-EEG-2...”和“Across 60 evaluation settings...”是实验结果。 不对,要注意不能增删改内容,所以小节标题可以是原文里的核心表述?不对,等下,规则说保留原文所有信息,不得增删改,那小节标题是不是要从原文里来?或者是不是不需要硬加小节?不对,看原文的结构,其实原文是一个连续的段落,但可能需要按内容划分?不对,再仔细看用户的要求:每个自然段落用

包裹,长文用

/

划分小节。那原文的纯文本是一个大的段落吗?不,原文的正文是连续的,但内容有几个部分:现有方法的局限、提出的新校准设置、实验协议、提出的GALA方法、实验结果。那可以划分小节,比如:

Neural decoding of visual experience from electroencephalography (EEG) has achieved strong performance in intra-subject settings, but current methods fail to generalize across individuals: on the standard 200-way benchmark THINGS-EEG-2, top-1 accuracy drops from above 80% within subjects to about 32% for an unseen subject. Adapting a pre-trained encoder to new subjects and sessions typically requires labeled data, a bottleneck whose collection time severely limits practical device deployment.

然后

提出的新校准设置

?不对,不能自己加内容?哦,不对,规则说不得增删改内容,那小节标题是不是不能自己编?那是不是不需要加小节?或者原文里的内容直接用

?不对,再看,原文的内容: 接下来的内容:“In a practical Brain-Computer Interface (BCI), however, the possible outcomes are known in advance and form a gallery of candidates, and the unlabeled EEG of the user accumulates as soon as the device is used. We propose a new setting, incremental transductive calibration, in which the decoder is calibrated regularly on all the unlabeled EEG received so far and on the known gallery, without any labels. This setting is closer to the conditions of a deployed BCI than the zero-shot and batch evaluations in current use.” 这部分可以用

,然后“We evaluate it with a protocol in which queries arrive one at a time, with unknown and possibly unequal frequencies, and each query is decoded before it is used for calibration.” 这部分是实验协议,然后“In this setting, we introduce GALA : GAllery-Level Alignment, a label-free calibration method which registers the unlabeled EEG embeddings onto the gallery through shrinkage whitening, hubness correction and a Sinkhorn--Procrustes alignment with query frequencies estimated from the buffer, updated after every query.” 这里GALA可以用?然后实验结果的两句用

。 不对,再整理: 首先,第一部分是现有方法的问题,用

包裹:

Neural decoding of visual experience from electroencephalography (EEG) has achieved strong performance in intra-subject settings, but current methods fail to generalize across individuals: on the standard 200-way benchmark THINGS-EEG-2, top-1 accuracy drops from above 80% within subjects to about 32% for an unseen subject. Adapting a pre-trained encoder to new subjects and sessions typically requires labeled data, a bottleneck whose collection time severely limits practical device deployment.

然后第二部分讲提出的新设置,用

增量传导校准设置

?不对,不能自己加标题?哦,等下,规则说“不得增删改内容”,那如果加小节标题的话,标题是不是属于新增内容?那是不是不需要加小节?那直接把每个自然段落用

?但原文是一个大段落,那怎么处理?不对,再看原文的表述,其实是连贯的,但可以按语义划分,不过不能新增内容,那是不是可以不用h2/h3?或者原文里的“incremental transductive calibration”是核心,那可以用h2吗?不对,等下,规则说“长文用

/

划分小节”,如果原文是长文的话可以划分,但是标题必须来自原文吗?比如

增量传导校准设置

是不是新增?不行,那可能不需要划分小节,直接用

分段? 等下再仔细看原文,原文的正文是连续的,没有分段,但我们可以按语义拆分段落,因为每个自然段落用

,那我们可以把不同的语义块拆成

,然后重要的内容用,比如GALA的定义? 哦,对,不能新增内容,所以小节标题不能自己编,那如果不需要小节的话,就直接用

拆分语义段落,然后重要的句子用。 现在来拆分: 第一段(现状与问题):

Neural decoding of visual experience from electroencephalography (EEG) has achieved strong performance in intra-subject settings, but current methods fail to generalize across individuals: on the standard 200-way benchmark THINGS-EEG-2, top-1 accuracy drops from above 80% within subjects to about 32% for an unseen subject. Adapting a pre-trained encoder to new subjects and sessions typically requires labeled data, a bottleneck whose collection time severely limits practical device deployment.

第二段(提出新设置的背景与内容):

In a practical Brain-Computer Interface (BCI), however, the possible outcomes are known in advance and form a gallery of candidates, and the unlabeled EEG of the user accumulates as soon as the device is used. We propose a new setting, incremental transductive calibration, in which the decoder is calibrated regularly on all the unlabeled EEG received so far and on the known gallery, without any labels. This setting is closer to the conditions of a deployed BCI than the zero-shot and batch evaluations in current use.

第三段(实验协议):

We evaluate it with a protocol in which queries arrive one at a time, with unknown and possibly unequal frequencies, and each query is decoded before it is used for calibration.

第四段(提出GALA方法):

In this setting, we introduce GALA : GAllery-Level Alignment, a label-free calibration method which registers the unlabeled EEG embeddings onto the gallery through shrinkage whitening, hubness correction and a Sinkhorn--Procrustes alignment with query frequencies estimated from the buffer, updated after every query.

第五段(实验结果):

On THINGS-EEG-2, with a gallery of 200 images, it raises the top-1 accuracy on a new subject from 28% to 67% over 53 minutes of use. Across 60 evaluation settings that vary the gallery size, the query frequencies and the number of repetitions averaged per query, GALA is the most accurate label-free method in every setting, 6.6 points above the best competing method on average at the end of the