内容纲要

in the case of  对于。。。而言
retain 保留
to the best of our knowledge 据我们所知
leverage,takes advantage of ,make best use of 充分利用
we are typically forced to use…  我们通常被迫使用
Traversal 遍历
in more pragmatic terms 用更实际的话
In its simplest incarnation 以最简单的形式
inserted as a plug-and-play module into  作为即插即用的模块插入到

intra-class commonality与 inter-class uniqueness
intra-class compactness和 inter-class discriminativeness.
inter:不同类之间  intra:同一类内部各部分之间。

in few-shot learning the two sets are exclusive 不同的
whilst 同时,而
In a nutshell 简而言之
alleviate this issue 缓解/解决这个问题
A important distinction to make is  一个重要的区别就是
hierarchy 层次 层级
We demonstrate the effectiveness of our 我们证明了我们的有效性
Our work goes beyond this by also…  我们的工作还超越了
looks to the query to determine… 查询query集来确定
most pertain to the task… 与任务最相关
As illustrated in Fig 如图所示
In practice 在实践中
we modify them to use our CTM as follows 如下所示
The contributions of this work are as follows:
The main contribution of this work is threefold:
a handful of labeled exemplars   几个 不多,极少数   模式,形式
FSL paradigm 模式
it is by no means trivial to。。。 这绝不是小事
To tackle this issue  为了解决这个问题
undergo 经历,通过
derive 导出,推出
critical obstacle 主要(关键)障碍
by two means 通过两种方式
extrapolate 预测,推测,推断
in an implicit way 以隐含的方式
In order to compensate for the scarcity of: 为了弥补。。。稀缺
in a unified way  以统一的方式
in a more principled way  以更合理的方式
give/provide as input to the GNN …  模型。。作为输入
prescribes to 按照/规定
promising solutions 有前途的解决方法
off-the-shelf 现成的
i.e.  即
As such 因此
It is not straightforward to  这并不容易
prone to 倾向于
on the fly 已经有现成的,直接拿来用。随手,直接做。
Hereafter 此后,随后
auxiliary 辅助的
alleviate over-fitting  缓和过拟合
an feed-forward manner 前馈方式
intermediate 中间的
Concretely ,To be concrete 具体来说
properties 性质
degenerate into 公式退化成
diverse hypothesis 许多假设(在。。。框架下,许多假设被提出Under the framework of meta-learning, a diverse hypothesis was made to build an efficient meta-learner)
substantial distribution relations  重要的/大量的
devise propose introduce  design  develop outline 设计,发明,提出
uni-modal 单模态
only a handful of existing works 只有少数现有的工作
there has been a recent resurgence of research interest in xxxx  再起
a linear classifier coupled with a deep feature extractor 耦合
two types of strategy resort to model the data distribution of novel category  采取,用来
the wrongly labeled instances may jeopardize the performance of the classifier  危害
To this end,In this pursuit  为此,因此
amend 修正
mimic a complete distribution 模拟完整分布
reawakened the research interest on 重新唤醒对。。。的兴趣
prominent objects 显著目标(数量多)rare instances
effortlessly 不费力,轻而易举
prevailing 流行的
per se 拉丁语,等于itself,要斜体
large-scale vision-language corpus 语料库
Ther is a trend towards A from B  有一种从B到A的趋势
we refer to XXXXnet as XXXX 我们把这种称为。。。
XXXX is sufficed xxxx已经足够了
Recent advances have leveraged different objective functions for xxx 最新进展利用了xxx 来 xxx
systematically 系统地,有条不紊的
take/draw inspiration from 从。。。得到灵感
alleviate overfitting,but not solve it 避免过拟合但是没由解决这个问题
allow novel examples to be rapidly assimilated, whist not suffering from catastorphic forgetting 同化/吸收
sensible 合理的
self-contained 自给自足,完整的,独立的
versatile:通用的
myriad small but related datasets:无数的
Due to the ubiquity of recent work, a unifying view is needed to understand and improve these methods.  普遍存在。
in effect 实际上
making it more amenable to classification tasks 更适合分类任务
without bells and whistles 没有花里胡哨
because the number of negative pairs grows quadratically w.r.t. the mini-batch size. 负样本对随着mini-batch乘方关系增加。
lookup and update 查询和更新
Despite their simplicity, these methods are very effective and often competitive with the state-of-the-art // simple yet effecive and competitive
A scalable solution 可扩展的方案
accuracy to quickly deteriorate  准确率恶化
catastrophic forgetting, fast, stable acquisition of new knowledge, learning with an imbalanced class labels, and fast learning during evaluation.  memory的优点
temporarily 暂时的
be scalable to 可拓展到
This compelling result along our thorough set of experiments on supervised and semi-supervised FSL.  令人信服的, 一整套实验。
to remedy this:  为了纠正这些。
few-shot classification has two main challenges: One is unseen classes, i.e., the non-overlap between training and test classes; The other is the low-data problem, i.e., very few labeled samples for the test unseen classes。
This is also formally called as a pictogram, pictogramme, pictograph, simply picto or icon. In this work, we interchangeably refer to an ideogram using the word “symbol” for simplicity.
compactly 简洁的
perceptron 感知器
substitute 替代
assess 评估
justified 正当的
opt for 采取
intractable 很难应付
surrogated 代替
pledge us to do  保证了我们可以做
numerator 分子
denominator 分母
The emergence of deep learning has advanced the image classification performance into an unprecedented level 出现, 史无前例的
The error rate on ImageNet has been halved and halved again,even approaching human-level performance.
impriented 印花的,加印的。
penultimate 倒数第二
properties 属性
Recent work has blurred the divide between triplet-based embedding training and softmax classification  消除了两者之间的鸿沟
to see this 为此
In stark contrast 最明显的不同。
Without loss of generality 不失一般性
intractable 棘手的
mutually-exclusive 互斥的
mitigate 缓和,减轻
compat 紧凑的
On one side of the spectrum  we take a different perspective.  从另一个角度
discrepancy 差异
obscures 隐蔽
the discrepancy of the implementation details among multiple few-shot learning algorithms obscures the relative performance gain.
granularity 粒度
arbitrary 任意的
be inline with 和xxx一致
a spectrum of 一系列
ensemble 合集
has significant adverse implications 有重大不利影响
conjecture 猜想
they suffer from the need to fine-tune on the target problem, which makes them less appealing to few-shot learning.
on par with 不相上下
surveillance video 监控视频
Anomaly Detection 异常检测
side-step the need for 回避需要
susceptible to 易受
Intriguingly 有趣的是

最后修改日期:2020年10月22日

作者

留言

撰写回覆或留言

发布留言必须填写的电子邮件地址不会公开。