“Racorn-k”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
（→‎Corn-k）
第30行： 第30行：
 
of the present market, where the similarity is measured by
 
of the present market, where the similarity is measured by
 
the Pearson correlation coefficient. This patten matching process
 
the Pearson correlation coefficient. This patten matching process
−
produces a set of similar periods, which we denote by E<sub>t</sub>. Then do a optimization following the idea
+
produces a set of similar periods, which we denote by C. Then do a optimization following the idea
−
of BCRP[] on E<sub>t</sub>.
+
of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are
 +
weighted to derive the ensemble-based portfolio.  
  
−
[[文件:Dnn-spk.png|500px]]
+
==Racorn-k==
  
−
However, the vanilla structure of Ehsan et al. performs rather poor compared to the i-vector counterpart. One reason is
+
At the t-th trading period, the CORN-K algorithm first selects
−
that the simple back-end scoring is based on average to derive the utterance-based representations (called d-vectors) , but
+
all the historical periods whose market status is similar to that
−
another reason is the vanilla DNN structure that does not consider much of the context and pattern learning. We therefore
+
of the present market, where the similarity is measured by
−
proposed a CT-DNN model that can learn stronger speaker features. The structure is shown below[1]:
+
the Pearson correlation coefficient. This patten matching process
 +
produces a set of similar periods, which we denote by C. Then do a optimization following the idea
 +
of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are
 +
weighted to derive the ensemble-based portfolio.
  
−
[[文件:Ctdnn-spk.png|500px]]
+
==Racorn(c)-k==
  
 +
At the t-th trading period, the CORN-K algorithm first selects
 +
all the historical periods whose market status is similar to that
 +
of the present market, where the similarity is measured by
 +
the Pearson correlation coefficient. This patten matching process
 +
produces a set of similar periods, which we denote by C. Then do a optimization following the idea
 +
of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are
 +
weighted to derive the ensemble-based portfolio.
  
−
Recently, we found that an 'all-info' training is effective for learning features. Looking back to DNN and CT-DNN, although the features
 
−
read from last hidden layer are discriminative, but not 'all discriminative', because some discriminant info can be also impelemented
 
−
in the last affine layer. A better strategy is let the feature generation net (feature net) learns all the things of discrimination.
 
−
To achieve this, we discarded the parametric classifier (the last affine layer) and use the simple cosine distance to conduct the
 
−
classification. An iterative training scheme can be used to implement this idea, that is, after each epoch, averaging the speaker
 
−
features to derive speaker vectors, and then use the speaker vectors to replace the last hidden layer. The training will be then
 
−
taken as usual. The new structure is as follows[4]:
 
  
  
−
[[文件:fullinfo-spk.png|500px]]
+
=Experiments on several datasets=
  
−
=Speech factorization=
 
  
 +
{| class="wikitable"
 +
!
 +
!
 +
!
 +
!
 +
!
 +
|-
 +
|
 +
|
 +
|
 +
|
 +
|
 +
|-
 +
|
 +
|
 +
|
 +
|
 +
|
 +
|-
 +
|
 +
|
 +
|
 +
|
 +
|
 +
|}
  
−
The short-time property is a very nice thing, which tells us it is possible to factorize speech signals. By factorization, we can achieve significant benefits:
 
  
−
A. Individual tasks can be largely improved, as unrelated factors have been removed.
+
=View the impact of risk-aversion=
−
 
+
−
B. Factors that are disturbs now becomes valuables things, leading to conditional training and collaborative training [5].
+
−
 
+
−
C. Once the factors have been separated, single factors can be manipulated, and reassemble these factors can change the signal according to the need.
+
−
 
+
−
D. It is a new speech coding scheme that leverage knowledge learned from large data.
+
−
 
+
−
 
+
−
Traditional factorization methods are based on probabilistic models and maximum likelihood learning. For example, in JFA, a linear Gaussian is assumed
+
−
for speaker and channel, and then a ML estimation is applied to estimate the loading matrices of each factor, based on a long duration of speech. Almost
+
−
all these factorizations share these features: shallow, linear, Gaussian, long-term segments.
+
−
 
+
−
We are more interested in factorization on frame-level, and plays not much assumption on how the factors are mixed. A cascaded factorization approach has been
+
−
proposed[6]. The basic idea is to factorize significant factors first,and then conditioned on the factors that have been derived. The architecture is as
+
−
follows, where we factorized speech signals into three factors: linguistic content, speaker trait, emotion. When factorizing each factor, supervised
+
−
learning is used. Note by this architecture, databases with different target labels can be used in a complementary way, which is different from
+
−
previously joint training approach that needs full-labelled data.
+
−
 
+
−
 
+
−
[[文件:deepfact.png|500px]]
+
−
 
+
−
 
+
−
=Speech reconstruction=
+
  
 
To verify the factorization, we can reconstruct the speech signal from the factors. The reconstruction is simply based on a DNN,
 
To verify the factorization, we can reconstruct the speech signal from the factors. The reconstruction is simply based on a DNN,
第100行： 第104行：
 
[[文件:fact-recover.png|500px]]
 
[[文件:fact-recover.png|500px]]
  
−
 
−
More recovery examples can be found [[dsf-examples|here]].
 
−
 
−
 
−
==Listen to the reconstruction==
 
−
 
−
We can listen to the wave for each factor, by using the original phase.
 
−
 
−
 
−
Original speech: [http://wangd.cslt.org/research/cdf/demo/CHEAVD_1_1_E02_001_worried.wav]
 
−
 
−
Linguistic factor: [http://wangd.cslt.org/research/cdf/demo/phone.wav]
 
−
 
−
Speaker factor: [http://wangd.cslt.org/research/cdf/demo/speaker.wav]
 
−
 
−
Emotion factor: [http://wangd.cslt.org/research/cdf/demo/emotion.wav]
 
−
 
−
Liguistic+ Speaker + Emotion: [http://wangd.cslt.org/research/cdf/demo/recovery.wav]
 
−
 
−
 
−
=Research directions=
 
−
 
−
* Adversarial factor learning
 
−
* Phone-aware multiple d-vector back-end for speaker recognition
 
−
* TTS adaptation based on speaker factors
 
  
  

2017年10月31日 (二) 02:00的版本

Project name

RACORN-K: RISK-AVERSION PATTERN MATCHING-BASED PORTFOLIO SELECTION

Project members

Yang Wang, Dong Wang, Yaodong Wang, You Zhang

Introduction

Portfolio selection is the central task for assets management, but it turns out to be very challenging. Methods based on pattern matching, particularly the CORN-K algorithm, have achieved promising performance on several stock markets. A key shortage of the existing pattern matching methods, however, is that the risk is largely ignored when optimizing portfolios, which may lead to unreliable profits, particularly in volatile markets. To make up this shortcoming， We propose a risk-aversion CORN-K algorithm, RACORN-K, that penalizes risk when searching for optimal portfolios. Experiment results demonstrate that the new algorithm can deliver notable and reliable improvements in terms of return, Sharp ratio and maximum drawdown, especially on volatile markets.


Corn-k

At the t-th trading period, the CORN-K algorithm first selects all the historical periods whose market status is similar to that of the present market, where the similarity is measured by the Pearson correlation coefficient. This patten matching process produces a set of similar periods, which we denote by C. Then do a optimization following the idea of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are weighted to derive the ensemble-based portfolio.

Racorn-k

At the t-th trading period, the CORN-K algorithm first selects all the historical periods whose market status is similar to that of the present market, where the similarity is measured by the Pearson correlation coefficient. This patten matching process produces a set of similar periods, which we denote by C. Then do a optimization following the idea of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are weighted to derive the ensemble-based portfolio.

Racorn(c)-k

At the t-th trading period, the CORN-K algorithm first selects all the historical periods whose market status is similar to that of the present market, where the similarity is measured by the Pearson correlation coefficient. This patten matching process produces a set of similar periods, which we denote by C. Then do a optimization following the idea of BCRP[] on C. Finally, the outputs of the top-k experts that have achieved the highest accumulated return are weighted to derive the ensemble-based portfolio.


Experiments on several datasets


View the impact of risk-aversion

To verify the factorization, we can reconstruct the speech signal from the factors. The reconstruction is simply based on a DNN, as shown below. Each factor passes a unique deep neural net, the output of the three DNNs are added together, and compared with the target, which is the logarithm of the spectrum of the original signal. This means that the output of the DNNs of the three factors are assumed to be convolved together to produce the original speech.

Fact-recover-dnn.png

Note that the factors are learned from Fbanks, by which some speech information has been lost, however the recovery is rather successfull.


View the reconstruction

Fact-recover.png


Reference

[1] Lantian Li, Yixiang Chen, Ying Shi, Zhiyuan Tang, and Dong Wang, “Deep speaker feature learning for text-independent speaker verification,”, Interspeech 2017.

[2] Ehsan Variani, Xin Lei, Erik McDermott, Ignacio Lopez Moreno, and Javier Gonzalez-Dominguez, “Deep neural networks for small footprint text-dependent speaker verification,”, ICASSP 2014.

[3] Lantian Li, Dong Wang, Yixiang Chen, Ying Shing, Zhiyuan Tang, http://wangd.cslt.org/public/pdf/spkfact.pdf

[4] Lantian Li, Zhiyuan Tang, Dong Wang, FULL-INFO TRAINING FOR DEEP SPEAKER FEATURE LEARNING, http://wangd.cslt.org/public/pdf/mlspk.pdf

[5] Zhiyuan Thang, Lantian Li, Dong Wang, Ravi Vipperla "Collaborative Joint Training with Multi-task Recurrent Model for Speech and Speaker Recognition", IEEE Trans. on Audio, Speech and Language Processing, vol. 25, no.3, March 2017.

[6] Dong Wang,Lantian Li,Ying Shi,Yixiang Chen,Zhiyuan Tang., "Deep Factorization for Speech Signal", https://arxiv.org/abs/1706.01777

取自“http://index.cslt.org/mediawiki/index.php?title=Racorn-k&oldid=29027”