Download presentation
Presentation is loading. Please wait.
Published byAda Paul Modified over 8 years ago
1
Dan Trout COSC 480
2
Knowledge Infusion (KI) The Guillotine Game Ottho (On the Tip of My Thought) Cognitive Units Knowledge Sources as CU Repositories Reasoning Mechanism Experiment Summary & Questions
3
“Process of providing a system with background knowledge that gives it a deeper understanding of the information it deals with.” The Matrix Web = source of world knowledge Humans offload cognitive processing to the Web
4
Italian game show Player given five words as clues Clues unrelated, each strongly linked to game’s unique solution
5
Clues ◦ Sin Forbidden fruit ◦ Newton Discovered gravity ◦ Doctor “Apple a day” ◦ Pie Obvious ◦ New York “Big Apple” Apple
6
We build Ottho’s knowledge through sources ◦ Encyclopedia (Italian Wikipedia) ◦ Italian dictionary ◦ Compound words – words that often go together ◦ Proverbs and aphorisms (Wikiquote) ◦ Descriptions of Italian movies (www.imdb.com)www.imdb.com ◦ Italian songs (www.onlylyrics.com)www.onlylyrics.com ◦ Book titles
7
Core of the system Strategy based on thought theory ◦ Humans’ long term memory encoded as CU ◦ Form interconnected network Represent each CU by two fields ◦ Head – words identifying concept ◦ Body – words describing CU Example ◦ CU = [artificial intelligence | intelligence computer science agent McCarthy reasoning…]
8
Ottho System Architecture
9
Different heuristics for each CU ◦ Encyclopedia Head = title, body = categories, bold words, section titles, page links ◦ Dictionary Head = lemma & synonyms, body = words in definition ◦ Compound words Head is empty, body = sequence of words in compound form ◦ Proverbs and aphorisms Head = author, body = words in quote ◦ Songs, movies, and books Head = author, body = words in title Extraction process = 743,193 CUs
10
Words and meanings stored in a network-like structure Spreading Activation Network (SAN) Network of searched nodes Building SAN ◦ M knowledge sources (KS 1,…,KS M ) = CU repositories ◦ CU repositories have five clues (k 1,…,k 5 ) ◦ Populate SAN by adding CUs related to clues ◦ For each clue k i, search performed in KS m (m=1,…,M ) to retrieve a list of relevant CUs.
11
Finally, we get M list of pairs CU j is the jth CU retrieved from KS m and w ij is the cosine similarity value between k i and CU j Example
12
Wikipedia Dictionary Wikipedia Dictionary
14
h = max. relevant CUs t = threshold for scoring words
15
Knowledge Infusion (KI) The Guillotine Game ◦ Ottho (On the Tip of My Thought) Cognitive Units Knowledge Sources as CU Repositories Reasoning mechanism Experiment
16
Giovanni, S. (2012). An Artificial Player for a Language Game. G. Marco de, L. Pasquale and B. Pierpaolo. 27: 36-43.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.