[Most-ai-contest] Please briefly summarize your current approach/implementation

[kysu]

Dear all,

 

Wish you all have a good New Year holiday!

 

To have a good score in the second/final round, we need to further raise our performance by improving the approach/model currently adopted. Please briefly summarize your current approach and then send it back to Dr. JJ Fan and me by Jan. 13 (Monday), the day before the next meeting. If your current approach (or an important component) just follows an existing approach, please give the associated reference. If you adopt a rule-based approach, just simply describe the property of those rules (e.g., adopting expression to catch lexicon/POS n-gram, etc.). Thanks for your help in advance.

 

KY

 

From: most-ai-contest-bounces at iis.sinica.edu.tw [mailto:most-ai-contest-bounces at iis.sinica.edu.tw] On Behalf Of 范正忠
Sent: Tuesday, December 31, 2019 11:22 AM
To: Most-ai Contest <Most-ai-contest at iis.sinica.edu.tw>
Subject: [Most-ai-contest] Fwd: We have entered the 2020科技大擂台 second/final round!

 

Dear all,

 

First we like to congratulate our team pass the first run and our hard work is worthy.

 

The next run (final run) will be held on 2020/4/8 and we still need your support to go forward the final run.

The next meeting will be 2020/1/14 12:30 and the agenda is listed as follows for your reference.

 

1. Discussion of our first run result (including the integration, implementation, and issues for future improvement) by jjfan

2. Description of your model implementation (including your model/method and which papers referenced in your model if you have)

3. Prof. Su will give an brief of some important concepts (NN-based, statistics-based, rule-based approaches) for your reference 

 

1.     BERT is very inefficient in learning an inference (even it is a simple one)

2.     Discrete-Symbol vs. Continuous-Embedding: Discrete-Symbols have no idea about the distance among them (even integers if each integer is regarded as a discrete symbol), and cannot merge their information.

3.     Rule/Statistics-based approaches have difficulty to handle synonyms (could be handled by WordNet), similar words, paraphrasing, implication, association.

4.     Rule/Statistics-based approaches have difficulty to handle context (they cannot merge them to keep the same dimensionality).

5.     NN-based approaches have difficulty to know if two lexicons/NEs are exactly the same

6.     NN-based approaches have difficulty to handle long-distance dependency/pattern (e.g., which one should be compared with which one, formatting pattern), aggregative features (e.g., counts, minimum/maximum, intersection, etc.), abstract features (e.g., P and H are the same regardless of their content)

7.     We should use NN-based approach to locate the regions, and then use rule-based approach to handle the boundary (which uses formatting regularity).

8.     We could use NN-based approach to locate/identify the roles, and then use rule-based approach to handle the inference (e.g., which one is the largest one; how many cities that he visited, etc.).

9.     Handle Topic/Focus (Zero anaphora, Omission, implicit implication): (1) Extract explicit topic/focus, (2) Use embedding to approximate topic/focus (e.g., use document-embedding to approximate topic, paragraph/sentence-embedding to approximate focus, RNN to feed in status of previous paragraph/sentence information to approximate zero-anaphora, omission).

 

4. The plan for the final run

 

Best,

jjfan

 

  _____  

From: "kysu" <kysu at iis.sinica.edu.tw <mailto:kysu at iis.sinica.edu.tw> >
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw <mailto:Most-ai-contest at iis.sinica.edu.tw> >
Cc: "范正忠" <jjfan at iis.sinica.edu.tw <mailto:jjfan at iis.sinica.edu.tw> >, "kysu" <kysu at iis.sinica.edu.tw <mailto:kysu at iis.sinica.edu.tw> >
Sent: Monday, December 30, 2019 8:58:52 AM
Subject: We have entered the 2020科技大擂台 second/final round!

 

Dear all,

 

According to Dr. JJ Fan, we have entered the second/final round (More details will be announced tomorrow, as Dr. JJ Fan will take off today to see a doctor). Thanks a lot for your great effort! We have made it. 

 

KY

 

 

From: 科技大擂台 [mailto:gitlab at speech.nchc.org.tw] 
Sent: Thursday, December 26, 2019 3:29 PM
To: kysu at iis.sinica.edu.tw <mailto:kysu at iis.sinica.edu.tw> 
Subject: Re: 2020_fgc_forum | 【官方公告】【2020科技大擂台 緊急通知】科技大擂台初賽日程調整通知 (#40)

 

 <https://speech.nchc.org.tw/fgc> 科技大擂台 commented: 

@cjlin <https://speech.nchc.org.tw/cjlin>  

12/26 1800 環境封閉直到賽後，正式比賽將於封閉存取之環境下進行，主辦單位不主動採取任何方式介入並啓動相關程式，請參賽團隊以自動化排程之機制完成二場正式賽。如有特殊狀況，主辦單位將會主動通知團隊

結果將於比賽後，擇期公布

— 
View it on GitLab <https://speech.nchc.org.tw/fgc/2020_fgc_forum/issues/40#note_330> . 
You're receiving this email because of your account on speech.nchc.org.tw. If you'd like to receive fewer emails, you can unsubscribe <https://speech.nchc.org.tw/sent_notifications/e87f5b0c792b3477d423bc515657a165/unsubscribe>  from this thread or adjust your notification settings. 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.iis.sinica.edu.tw/pipermail/most-ai-contest/attachments/20200101/081a2415/attachment-0001.html>