CS 6601 Artificial Intelligence: Pattern Recognition Through Time

Section A

Dynamic Time Warping is a robust algorithm for determining similarity between two time-series sequences and can be applied to a large range of problems in the fields of shape recognition, handwriting recognition, and speech recognition. It allows for more flexibility than the Euclidean distance metric, and is used on data that has a large amount of variance. See the figure below for a visual depiction of how these two algorithms compare two similar time-series by comparing individual data points. Note that the Dynamic Time Warping algorithm matches data points that we would consider to be most suitable for comparison, while the Euclidean matching algorithm is a bit more naive.

One modern application of this algorithm is the classification of animal speech, in particular the classification of birdsongs. In this problem you’ll work through the details of such a classification system.

A young bird learning birdsongs can be compared to a toddler learning their first language. Initially the bird experiments with vocal abilities without organization, but through observation of parents and other members of the local community the bird slowly refines its song until it can reproduce the tune near-perfectly. Due to the reliance on learning from other local birds, different communities of birds of the same species can have slightly differing birdsongs. These locality and age factors can make it difficult to accurately recognize a bird’s species from it’s birdsong, but one successful method of solving this problem is by applying Dynamic Time Warping.

To construct a classification system, researchers first collect a large amount of data concerning birdsongs across many species, ages, and localities and construct a representative database. For this simplified problem, our database is comprised of the two species below: a yellow wagtail and a skylark. The waveforms show the typical form of the birdsong, while the time-series data represents the measurements that are actually in our database.

When a new birdsong from an unknown bird is recorded, the Dynamic Time Warping algorithm is applied to compare the new reading against birdsongs in the database, and the most similar bird is used as the classification label.

How do we actually perform the algorithm? Given two time series sequences A and B, a matrix of size (|A| x |B|) is initialized, and each element γ(  i,j   ) is computed using the following formula:

Fig 3: Application on example sequences A and B

Each element γ(  i,j   ) represents the total DTW distance if elements , are matched together. The most important value in this matrix is at the very top right, the value that indicates the total distance when all elements of both sequences have been matched; the total distance between A and B. To find which data points have been matched together, also known as the warping path or global alignment, you can work the algorithm backwards. Starting at the top right element, follow the neighbor that you used to calculate that element until you reach the bottom left. Given a tie in neighboring values, move diagonally, and continue to move diagonally until the tie is resolved. Every pair (i,j) in this path is the optimal pairing used to find the total distance.

Fig 4: Example of Warping Path

A1. Recorded birdsong is given to you with a frequency time-series sequence of:

1. Fill the table below as shown in Fig. 3 to perform the Dynamic Time Warping algorithm and find the total distance between the new time series and our database entry for the yellow wagtail. (3 points)

1. Find the Warping Path of the above computation and show which pairings are in our warping path by finding the path as shown in Fig. 4, and listing which data points from our Y axis are paired with the data points from our X axis. (1 point)

1. Find the Dynamic Time Warping distance between the new time series and our database entry for the skylark. (3 points)

1. Find the Warping Path of the above computation and show which pairings are in our warping path by finding the path as shown in Fig. 4, and listing which data points from our Y axis are paired with the data points from our X axis. (1 point)

Section B: Sakoe-Chiba Bands

This naive implementation of DTW has a time complexity of   (   ·   ) and a space complexity of   (   ·   ), where N and M are the lengths of the two time sequences. This is one of the most cited reasons to not use DTW. Another common reason is that the algorithm may be too flexible. For example, it may be realistic to match one datapoint from A to three data points from B, but it would almost certainly be inaccurate to match one datapoint from A to half the data points from B. Take a look at the example paths below, where the green path is acceptable and the red path is not very useful. Sakoe-Chiba bands are a prevalent solution to both these issues. In this section, you are going to compute the distance between two time series using this technique.

Fig 5: Comparison of DTW Paths

The path directly from the bottom left to top right of the above matrix represents a strict pairing where each point from A can be paired with only one point from B. This is inflexible, and a naive DTW allows us to disregard this constraint. However, Sakoe-Chiba bands impose the constraint that any element on the optimal path cannot be too far from this strict pairing. Below are examples of Sakoe-Chiba bands of degree N, where we will only evaluate the matrix elements a distance N away from the strict pairing line. Examine the examples below, where the gray boxes are disregarded in DTW calculations:

Fig 6: Visualization of Sakoe-Chiba bands

B.1 Using the same time-series data used in the previous problems, [8,8,6,7,5,3,0,9,2]:

1. Find the Dynamic Time Warping distance to our database entry for the yellow wagtail using the Sakoe-Chiba bands shown below. (3 points)

1. Find the Warping Path of the above computation and show which pairings are in our warping path by finding the path as shown in Fig. 4, and listing which data points from our Y axis are paired with the data points from our X axis. (1 point)

1. Find the Dynamic Time Warping distance to our database entry for the skylark using the Sakoe-Chiba band shown below. (3 points)

1. Find the Warping Path of the above computation and show which pairings are in our warping path by finding the path as shown in Fig. 4, and listing which data points from our Y axis are paired with the data points from our X axis. (1 point)

Answer in Detail

i
Solved by qualified expert

58 More Pages to Come in This Document. Get access to the complete answer.

Assignment

Essay: 2 Pages, Deadline:
3 days

Statistics

Home Work: 3 Pages, Deadline:
11 hours

Civil Law

Home Work: 3.6 Pages, Deadline:
4 hours

Economics

Home Work: 1.2 Pages, Deadline:
2 hours

Management

Assignment: 5 Pages, Deadline:
5 days

Maths

Home Work: 1 Page, Deadline:
54 minutes

Maths

Home Work: 1 Page, Deadline:
1 hour

Maths

Home Work: 1 Page, Deadline:
1 hour

Finance

Essay: 4 Pages, Deadline:
7 days

Nursing

Assignment: 1 Page, Deadline:
4 days

Nursing

Assignment: 12 Pages, Deadline:
5 days

Maths

Course Work: 0.4 Pages, Deadline:
2 hours

Marketing

Assignment: 3 Pages, Deadline:
3 days

Electronics

Essay: 1 Page, Deadline:
7 days

Biology

Programming: 4 Pages, Deadline:
1 day

Statistics

Programming: 16 Pages, Deadline:
4 days

Biology

Home Work: 5 Pages, Deadline:
5 days

Civil Law

Home Work: 10 Pages, Deadline:
8 days

Assignment

Assignment: 4 Pages, Deadline:
14 days

Organizational Behavior

Essay: 7.2 Pages, Deadline:
2 days

Very quick done and professional job. Helped me to gain good marks. Easy to read as answers given very clear.

User ID: 5***21 Blanchardstown, Ireland

Thank you very much… YOU ARE A GREAT WRITER ! You provide this answer and saved my life !

User ID: 8***32 Blanchardstown, Sri Lanka

Thanks for the time response and helped me in successfully submitting the assignment. Thanks for this time

User ID: 8***31 United Kingdom, Great Britain

Thanks for helping at short notice. Great quality, in the short time provided. All the best

User ID: 8***61 United Kingdom, Great Britain

This report addresses the assignment brief very well, with much detail on Unilever online presence. The recommendation are good, and more detail/ less …

User ID: 8***01 England, Great Britain

Super quick response time, great detail and construction of the solution. Many thanks for your help in this assignment. All the best

User ID: 8***61 United Kingdom, Great Britain

The new answer seems correct and is well explained in detail. Thank you for reviewing the work and for your help on this assignment. All the best.

User ID: 8***61 United Kingdom, Great Britain

Very good solution, well explained in detail. Thank you very much for your help on this assignment. All the best.

User ID: 8***61 United Kingdom, Great Britain

Awesome work! I am fully satisfied with the quality of the content. The writers worked hard to deliver such a deep and wide content. Keep it up.

User ID: 8***46 Batu Caves, Malaysia

This powerpoint was neat and concise. The expert definitely followed through with what was asked.

User ID: 6***42 Batu Caves, United States

well written assignment. good references and all done in a timely manner. pricing was good and support available when needed

User ID: 6***93 United Kingdom, Great Britain

Great, very clear, thank you very much For your help on this assignment. All the best

User ID: 8***61 United Kingdom, Great Britain

This website has been absolutely amazing and been very helpful and beneficial to my progress in my program.

User ID: 5***18 Mobile, United States

seems good (ok i am only writing this because of 80 characters, what more that seems good do you want?)

User ID: 8***00 Mobile, France

The effort and work they do is excellent and they have a great speed to complete tasks, I will always cooperate with them

User ID: 8***67 Belfast, Great Britain

the assignment is good which gave me a good marks and feedback i got from uni is good attempt

User ID: 6***66 United Kingdom, Great Britain

good and fast will use again helped me understand more I was able to incorporate the work done with mine

User ID: 8***83 United Kingdom, Great Britain

written very well, appreciate the help with writing this assignment. will definitely go through you this service again. very happy

User ID: 7***52 Brisbane, Australia

awesome work was very well written and covered all basics of the assignment. appreciate the help.

User ID: 7***52 Brisbane, Australia

The content was great and right on the mark. Very comprehensive and descriptive.

User ID: 8***23 Brisbane, Australia