The document summarizes a research paper that compares the performance of MLP-based models to Transformer-based models on various natural language processing and computer vision tasks. The key points are:
1. Gated MLP (gMLP) architectures can achieve performance comparable to Transformers on most tasks, demonstrating that attention mechanisms may not be strictly necessary.
2. However, attention still provides benefits for some NLP tasks, as models combining gMLP and attention outperformed pure gMLP models on certain benchmarks.
3. For computer vision, gMLP achieved results close to Vision Transformers and CNNs on image classification, indicating gMLP can match their data efficiency.
プレゼン・ポスターで自分の研究を「伝える」 (How to do technical oral/poster presentation)Toshihiko Yamasaki
MIRU2020若手プログラム招待講演のスライドを一般公開用にアレンジしたものです。日本語で書かれています。下記の点にご注意ください
・セリフが伴ってないので内容は限定的です
・著作権等に配慮しているので中身は結構無味乾燥です。
This is an arranged version of my invited talk at MIRU 2020 young researchers' forum. This is written in Japanese.
This document summarizes a presentation about variational autoencoders (VAEs) presented at the ICLR 2016 conference. The document discusses 5 VAE-related papers presented at ICLR 2016, including Importance Weighted Autoencoders, The Variational Fair Autoencoder, Generating Images from Captions with Attention, Variational Gaussian Process, and Variationally Auto-Encoded Deep Gaussian Processes. It also provides background on variational inference and VAEs, explaining how VAEs use neural networks to model probability distributions and maximize a lower bound on the log likelihood.
The document summarizes a research paper that compares the performance of MLP-based models to Transformer-based models on various natural language processing and computer vision tasks. The key points are:
1. Gated MLP (gMLP) architectures can achieve performance comparable to Transformers on most tasks, demonstrating that attention mechanisms may not be strictly necessary.
2. However, attention still provides benefits for some NLP tasks, as models combining gMLP and attention outperformed pure gMLP models on certain benchmarks.
3. For computer vision, gMLP achieved results close to Vision Transformers and CNNs on image classification, indicating gMLP can match their data efficiency.
プレゼン・ポスターで自分の研究を「伝える」 (How to do technical oral/poster presentation)Toshihiko Yamasaki
MIRU2020若手プログラム招待講演のスライドを一般公開用にアレンジしたものです。日本語で書かれています。下記の点にご注意ください
・セリフが伴ってないので内容は限定的です
・著作権等に配慮しているので中身は結構無味乾燥です。
This is an arranged version of my invited talk at MIRU 2020 young researchers' forum. This is written in Japanese.
This document summarizes a presentation about variational autoencoders (VAEs) presented at the ICLR 2016 conference. The document discusses 5 VAE-related papers presented at ICLR 2016, including Importance Weighted Autoencoders, The Variational Fair Autoencoder, Generating Images from Captions with Attention, Variational Gaussian Process, and Variationally Auto-Encoded Deep Gaussian Processes. It also provides background on variational inference and VAEs, explaining how VAEs use neural networks to model probability distributions and maximize a lower bound on the log likelihood.
The slides of Artificial Intelligence and Entertainment Science (AIES) Workshop 2021 Keynote lecture
https://aies.info/program/
Empathic Entertainment in Digital Game
A digital game give a unique experience to a user. AI system in Digital game consists of three kinds of AI such as Meta-AI, Character AI, and Spatial AI. Game experience is formed by them. Meta-AI keeps watching a status of game and controlling characters, objects, terrain, weather and so on dynamically to make many dramatic and empathic situations in a game for users. Character AI is a brain of an autonomous game character to make a decision by itself, but sometimes it acts to achieve a goal issued from Meta-AI. Spatial AI analyses a terrain and abstracts its features to communicate them to Meta-AI and Character-AI. They can make their intelligent decisions by using specific terrain and environment features. The AI system is called MCS-AI dynamic cooperative model (Meta-AI, Character AI, and Spatial AI dynamic cooperative model). In the lecture, I will explain the system by showing some cases of published digital games.
The document discusses the differences between making a microwave and creating artificial intelligence. It explores how intelligence may have common principles across different animals and how studying biology can help understand intelligence and realize it in computers and robots. It also discusses approaches to building AI through engineering as well as understanding what intelligence is through philosophy and science. Finally, it discusses game engines and their role in simulating physical, chemical, economic, social and biological rules to create virtual worlds.
100. A-Consciousness に関する3つのアイデア
(1) 黒板モデル =ブラックボード・アーキテクチャ
(Blackboard Architecture)
(2) GWT = Global Workspace Theory
(Baar, 1988)
(3) MDM = Multiple Draft Model
(Dennett, 1991)
3つのアイデアをかけあわせる
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
102. Global Workspace Theory (GWT)
注意の焦点
Focus of Attention
(スポットライト)
舞台裏の人々=ディレクター、シーンデザイナー、など。
コンテキストの生成とコントロール(舞台裏)
ワーキングメモリ
(Scene,Stage)
専門
プロセッサー
(観客)
ブロードキャスト
テンポラリー
な連携
Arrabales, R. Ledezma, A. and Sanchis, A. “Towards the Generation of Visual Qualia in Artificial Cognitive Architectures”. (2010) http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
103. CERA-CRANIUM認識モデル
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
104. 単純な認識 Single Percept が
得た情報から
さらにProcessor が
解釈した情報を、
同じ Workspaceに書き込む
CERA-CRANIUM認識モデル
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
105. CERA-CRANIUM認識モデル
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
106. CORE Layer は、Physical Laryer 、Mission Layer のうちで、
どの認識を生成するかを決定するコマンドを投げる。
CERA-CRANIUM認識モデル
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
154. S
(Body)
O
(Object)
Action: A
O’
Sense: p
S’
Action: A’
Sense: p’
O’’S’’
Action: A’’
Sense: p’’
O’’S’’
Action: A’’
Sense: p’’
R R
R R
R R
f’
f
f’’
Sequence of
Self
Sequence of
Object
Sequence of
Action
“Self” is a sequence of selfs. “Object” is a sequence of objects.
155. S
(Body)
w
(World)
Action: A
w’
Sense: p
S’
Action: A’
Sense: p’
w’’S’’
Action: A’’
Sense: p’’
w’’S’’
Action: A’’
Sense: p’’
R R
R R
R R
f’
f
f’’
Sequence of
Self
Sequence of
World
Sequence of
Action
“Self” is a sequence of selfs. “World” is a sequence of worlds.
知能は身体、世界、その間の関係(アクション)を多層的に捉えている。
多層的な表現の集合がオブジェクトであり、世界であり、アクション。
世界表現身体表現
(自己表現)
行動表現
156. S
(Body)
O
(Object)
Action: f
O’
Sense: p
S’
Action: f’
Sense: p’
O’’S’’
Action: f’’
Sense: p’’
R
R
R
R
Sequence of
Self
“Self” is a sequence of selfs. “Object” is a sequence of objects.
Vector to move
Vector to ent
Dude, Where's My Warthog: From Pathfinding to General Spatial Competence,
D. Isla, Invited talk, Artificial Intelligence and Interactive Digital Entertainment (AIIDE) 2005
http://naimadgames.com/publications.html
157. S
(Body)
O
(Object)
Action: f
O’
Sense: p
S’
Action: f’
Sense: p’
O’’S’’
Action: f’’
Sense: p’’
R
R
R
R
“Self” is a sequence of selfs. “Object” is a sequence of objects.
Killzone 2 Multiplayer Bots
Remco Straatman, Tim Verweij, Alex Champandard | Paris Game/AI Conference 2009, Paris, June 2009
http://www.guerrilla-games.com/publications.html
158. S
(Body)
O
(Object)
Action: f
O’
Sense: p
S’
Action: f’
Sense: p’
O’’S’’
Action: f’’
Sense: p’’
R
R
R
R
“Self” is a sequence of selfs. “Object” is a sequence of objects.
Handling Complexity in the Halo 2 AI, D. Isla, GDC 2005
Dude, Where's My Warthog: From Pathfinding to General Spatial Competence,
D. Isla, Invited talk, Artificial Intelligence and Interactive Digital Entertainment (AIIDE) 2005
http://naimadgames.com/publications.html
159. S
(Body)
O
(Object)
Action: f
O’
Sense: p
S’
Action: f’
Sense: p’
O’’S’’
Action: f’’
Sense: p’’
R
R
R
R
“Self” is a sequence of selfs. “Object” is a sequence of objects.
Handling Complexity in the Halo 2 AI, D. Isla, GDC 2005
Dude, Where's My Warthog: From Pathfinding to General Spatial Competence,
D. Isla, Invited talk, Artificial Intelligence and Interactive Digital Entertainment (AIIDE) 2005
http://naimadgames.com/publications.html
物事を多層的に認識する
220. いろいろな知識表現
事実表現(信頼度表現)
意味ネットワーク
敵表現リスト
依存グラフ ルールベース表現
世界表現
Griesemer,J, "The Illusion of Intelligence: The Integration of AI and Level Design in Halo", 2002
http://www.bungie.net/images/Inside/publications/presentations/publicationsdes/design/gdc02_jaime_griesemer.pdf
Agent Architecture Considerations for Real-Time Planning in Games (AIIDE 2005)
http://web.media.mit.edu/~jorkin/AIIDE05_Orkin_Planning.ppt
Beyond Behavior: An Introduction to Knowledge Representation, D. Isla, P. Gorniak, AI Summit GDC 2009
http://naimadgames.com/publications.html
221. いろいろな世界表現
ナビメッシュ-ウェイポイント
階層表現
LOS マップ
戦術マップクラスタリング
敵配位マップ テリトリー表現
Tactical Point System
Halo2Killzone
Killzone2Halo Assassin’s Creed
Left 4 Dead
Alex J. Champandard, Remco Straatman, Tim Verweij, "On the AI Strategy for KILLZONE 2's Bots”
http://aigamedev.com/open/coverage/killzone2/
Damian Isla,"Building a Better Battle: HALO 3 AI Objectives",
http://www.bungie.net/inside/publications.aspx
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment
Conference , http://www.valvesoftware.com/publications.html
225. オブジェクト表現
これが車である
この方向に押せば動く
Dude, Where's My Warthog: From Pathfinding to General Spatial Competence, D. Isla, Invited talk, Artificial Intelligence and Interactive Digital Entertainment (AIIDE) 2005
http://naimadgames.com/publications.html