论文的表达

插入语的位置

插入语	用法
of course	The disadvantage, of course, is that this is computationally very expenisve when we need to esitmate multiple parameters
e.g.	use of proposal distributions (e.g. rejection and importance sampling).

一些表达

原	用法
Under certain conditions	在特定条件下
we relax the condition
this assumption is violated
we can skip the first step and proceed with the calculation of the mean vecs	进行，执行

• These studies assume the vectors of the different channels to be jointly sparse. In this paper, we relax the condition.
• Numerous efforts have been made to counter the intra-class varaiability by manually designing low-level features for classification tasks at hand. Representative examples are

一些单词的学习

• comprise

  vt. 包含，由…组成

  In this work, we propose a very simple deep learning network for image classification which comprises only the very basic data processing components:

表达逻辑顺序

• it is followed by

算法的描述

Given an initial guess for θ with positive probability of being drawn, the Metropolis-Hastings algorithm proceeds as follows

- Choose 
- Calculate
- If
- Repeat the earlier steps

形容词

Image classification based on visual content is a very challenging task, largely because there is usually large amount of intra-class variability, arising from …

• 中英文互译时，有时需要改变原文中的句式结构，有时需要转换表达方式。这种结构以及表达方式的转换是鼓励的，当沿着原文的结构或者含义进行翻译出现障碍时，应当尝试转换。

  1. we can maximize the likelihood (or, equivalently, minimize the loss) to find θ within a frequentist paradigm.

  2. 施加影响：
     to exert influence on …
     It’s clear on examination that the outliers are exerting a disproportionate influence on the fit.

  3. The variety of possible loss functions is quite literally infinite, but one relatively well-motivated option is the Huber loss.

  4. A and B are C respectively：
     Ridge regression and the lasso are regularized versions of least squares regression using ℓ2 and ℓ1 penalties respectively, on the coefficient vector.

  5. account for：解释
     The Bayesian approach to accounting for outliers generally involves modifying the model so that the outliers are accounted for. For this data, it is abundantly clear that a simple straight line is not a good fit to our data. So let’s propose a more complicated model that has the flexibility to account for outliers. One option is to choose a mixture between a signal and a background:
     

     P({xi},{yi},{ei}|θ,{gi},σB)=gi2πe2i−−−−√exp[−(y^(xi|θ)−yi)22e2i]+1−gi2πσ2B−−−−−√exp(−(y^(xi|θ)−yi)22σ2B)
     
     What we’ve done is expanded our model with some nuisance parameters: {gi} is a series of weights which range from 0 to 1 and encode for each point i the degree to which it fits the model. gi=0 indicates an outlier, in which case a Gaussian of width σB is used in the computation of the likelihood. This σB can also be a nuisance parameter, or its value can be set at a sufficiently high number, say 50.

  6. 足够高：
     its value can be set at a sufficiently high number, say 50.

  7. tradeoff
     there is a tradeoff between query complexity and query stability.

  8. 细节保持
     detail-preserving，fine detail-preserving

  9. Dissimilarity Measure or Similarity Measure

  10. by optimizing an objective function that includes two terms: one that measures the signal reconstruction error and another that measures the sparsity.

  11. suffer from multiple local minima,

  12. The proposed approach outperforms the standard discriminative methods and the standard sparse representation in the case of corrupted signals.

  13. The problem solved by the sparse representation is to search for the most compact representation of a signal in terms of linear combination of atoms in an overcomplete dictionary.
      （as for，in terms of，as far as concerned）

      In terms of quality, A is better than B.
      As far as quality is concerned, A is better than B.
      

  14. The rest of this paper is organized as follows。

  15. The rest of this paper is organized as follows. Section 2 reviews the problem formulation and solution for the standard sparse representation. Section 3 discusses the motivations for proposing SRSC by analyzing the reconstructive methods and discriminative methods for signal classification. The formulation and solution of SRSC are presented in Section 4. Experimental results with synthetic and real data are shown in Section 5 and Section 6 concludes the paper with a summary of the proposed work and discussions.

  16. Inspired by this idea.

  17. Lets Assume an arbitrary Markov Chain P with infinite states on [0,1] having transition Matrix Q such that（满足） Qij=Qji= All entries in Matrix.

  18. One advantage of this is that the prior does not have to be conjugate (although the example below uses the same beta prior for ease of comaprsion), and so we are not restricted in our choice of an approproirate prior distribution.

      说的是 P(X) 的计算：
      

      p(X)=∫d(θ∗)p(θ∗)p(X|θ∗)
      
      使用数值积分（numerical integration）的方式（on a grid of values for θ ）进行计算：
      
      p(X)=∑θ∗p(θ∗)p(X|θ∗)

  19. All ocde will be built from the ground up to ilustrate what is involved in fitting an MCMC model, but only toy examples will be shown since the goal is conceptual understanding.