模型介绍

Logistic 分布

Logistic regression分布是一种二分类法则，其分布函数和密度函数为：

其中$\mu$表示位置参数，$\gamma>0$为形状参数，我们可以看以下形状特点：

C1-Orientation

1. Attendance & Reflection journal 20%
2. Written assignment 30%
3. academic oral presentations 20%
4. Scientific paper analysis 30%

Academic Style

No phrasal verbs: Go up/look up

No personal pronouns: I, we, opinion

No contractions: It’ll: It will

No Questions: so why did the project fall?

No personal adverbs: Surprisingly

No vague language: And so on/ etc.

No basic language: like: For example/thing: factor/ lots of: significant amount/amazing: Important

If our environment is not protected, our children will suffer the most.

If the environment is not protected, children will suffer the most.

I believe that we need to address these problems.

These problems need to be addressed.

<–more–>

Background

1. Previous methods model about Pedestrian trajectory prediction only the pairwise interactions between pedestrians, which not only oversimpfilies the interactions among pedestrians but also is computationally inefficient.
2. Other methods are limited by the quality of manually designed features.

Contribution

StarNet

There is two parameters in StarNet. The centreal part of StarNet is the hub network, which produces a reprensetation r of the interactions among pedestrians.

2015 Aggregating Deep Convolutional Features for Image Retrieval

2017 Fine-tuning CNN Image Retrieval with No Human Annotation

Background

Training of CNNs, either from scratch or fine-tuning, requires a large amount of annotated data, where a high quality of annotation is often crucial. In

Main contribution

1. fine-tune CNNs—– Reconstructed 3D models obtained CNN descriptor whitening discriminatively learned from the same training data outperforms commonly used PCA whitening.
2. Whitening: Traditonal whitening performed on short representations is, in some cases, unstable. We propose to learn the Whitening through the same training data. Performing whitening as a post-processing step is better and much faster to train compared to learning it end-to-end.
3. Generalized-Mean (GeM) pooling layer that generalizes max and average pooling and show that it boosts retrieval performance.
4. In addition, we propose a novel α-weighted query expansion that is more robust compared to the standard average query expansion technique widely used for compact image rep- resentations.
5. Applying the proposed method to the VGG network achieves state-of-the-art performance on the standard benchmarks: Oxford Buildings, Paris, and Holidays datasets.