Filters








999 Hits in 6.2 sec

Predicting What You Already Know Helps: Provable Self-Supervised Learning [article]

Jason D. Lee, Qi Lei, Nikunj Saunshi, Jiacheng Zhuo
2021 arXiv   pre-print
Self-supervised representation learning solves auxiliary prediction tasks (known as pretext tasks) without requiring labeled data to learn useful semantic representations.  ...  predicting this known information helps in learning representations effective for downstream prediction tasks.  ...  However, the underlying principles of self-supervised learning are mysterious since it is a-priori unclear why predicting what we already know should help.  ... 
arXiv:2008.01064v2 fatcat:ulpr5splhzft7bbdkjubwvxwb4

A Mathematical Exploration of Why Language Models Help Solve Downstream Tasks [article]

Nikunj Saunshi, Sadhika Malladi, Sanjeev Arora
2021 arXiv   pre-print
task of next word prediction and text classification?  ...  This paper initiates a mathematical study of this phenomenon for the downstream task of text classification by considering the following questions: (1) What is the intuitive connection between the pretraining  ...  We use the LogisticRegressionCV class from the scikit-learn package to fit linear classifiers to all fixed features (i.e., no finetuning).  ... 
arXiv:2010.03648v2 fatcat:ewo4ad423ndazjfyuea47g5t6a

Constructionist Steps Towards an Autonomously Empathetic System [article]

Trevor Buteau, Damian Lyons
2018 arXiv   pre-print
Thus, our system uses a user-dependent method of analysis and relies heavily on contextual information to make predictions about what subjects are experiencing.  ...  Prior efforts to create an autonomous computer system capable of predicting what a human being is thinking or feeling from facial expression data have been largely based on outdated, inaccurate models  ...  based on what you know about the context in which these feelings are occurring.  ... 
arXiv:1808.00981v1 fatcat:64i5lobmsncehepr737byw7k7q

About the Essence of Intelligence – Will Artificial Intelligence (Ever) Cover Human Intelligence? [chapter]

Hannu Jaakkola, Bernhard Thalheim, Jaak Henno
2022 Frontiers in Artificial Intelligence and Applications  
Abilities related to intelligence cover ability to acquire and apply knowledge and skills, as well as ability to learn.  ...  All the time a lot of discussion about intelligence of these systems has been going on – are the AI based systems and robots intelligent, what is the difference of human and machine intelligence, etc.  ...  This has created a new research direction: Explainable AI (XAI) -developing tools and frameworks to help you understand and interpret predictions made by your machine learning models [3; 31].  ... 
doi:10.3233/faia210475 fatcat:spntubdzm5anhcnssjolcmhre4

Self-supervised Learning from a Multi-view Perspective [article]

Yao-Hung Hubert Tsai, Yue Wu, Ruslan Salakhutdinov, Louis-Philippe Morency
2021 arXiv   pre-print
As a subset of unsupervised representation learning, self-supervised representation learning adopts self-defined signals as supervision and uses the learned representation for downstream tasks, such as  ...  Our theoretical framework paves the way to a larger space of self-supervised learning objective design.  ...  Predicting what you already know helps: Provable self-supervised learning. arXiv preprint arXiv:2008.01064, 2020.  ... 
arXiv:2006.05576v4 fatcat:slhsl2b7lbcepnw56ehw7oc3c4

Zipfian environments for Reinforcement Learning [article]

Stephanie C. Y. Chan and Andrew K. Lampinen and Pierre H. Richemond and Felix Hill
2022 arXiv   pre-print
memory system and self-supervised learning objectives can all influence an agent's ability to learn from uncommon experiences.  ...  To understand this failure better, we explore how different aspects of current approaches may be adjusted to help improve performance on rare events, and show that the RL objective function, the agent's  ...  Self-supervised learning.  ... 
arXiv:2203.08222v1 fatcat:vn2kaznzlngwpg6jndabipz6qa

PowerPlay: Training an Increasingly General Problem Solver by Continually Searching for the Simplest Still Unsolvable Problem

Jürgen Schmidhuber
2013 Frontiers in Psychology  
New skills may (partially) re-use previously learned skills.  ...  solves all previously learned tasks plus the new one, while the unmodified predecessor does not.  ...  already knows about the (external or internal) world's repetitive patterns.  ... 
doi:10.3389/fpsyg.2013.00313 pmid:23761771 pmcid:PMC3675324 fatcat:b3kabrx2ynggvleapkff4n62w4

Critical Thinking in the Management Classroom: Bloom's Taxonomy as a Learning Tool

Nicholas Athanassiou, Jeanne M. McNett, Carol Harvey
2003 Journal of Management Education  
Because we are concerned with what seems to be a universal problem, that is, that many students do not think critically and do not integrate what they are learning with what they already know, the focus  ...  This helped them out of the trap of not recognizing what they do not know.  ... 
doi:10.1177/1052562903252515 fatcat:7iczcnpyybcktbaiqzr54nvs3m

One Decade of Universal Artificial Intelligence [chapter]

Marcus Hutter
2012 Atlantis Thinking Machines  
For the first time, without providing any domain knowledge, the same agent is able to self-adapt to a diverse range of interactive environments.  ...  For instance, AIXI is able to learn from scratch to play TicTacToe, Pacman, Kuhn Poker, and other games by trial and error, without even providing the rules of the games.  ...  What would be very interesting to show is that logic is an emergent phenomenon, i.e. that AIXI learns to reason logically if/since this helps collect reward.  ... 
doi:10.2991/978-94-91216-62-6_5 fatcat:scdb4kilsnc55f5fxk3qurtd7i

Recommendations as Treatments

Thorsten Joachims, Ben London, Yi Su, Adith Swaminathan, Lequn Wang
2021 The AI Magazine  
This article explains how these techniques enable unbiased offline evaluation and learning despite biased data, and how they can inform considerations of fairness and equity in recommender systems.  ...  We pause here to note that traditional metrics used for supervised learning-such as the accuracy of click prediction-do not account for these inherent biases.  ...  However, we can provide technical answers for how to design systems that guarantee certain trade-offs, help reason about what can and what cannot be implemented, and help understand the dynamics of such  ... 
doi:10.1609/aimag.v42i3.18141 fatcat:hdyi4nadijgp3fpieqojib5pfq

One Decade of Universal Artificial Intelligence [article]

Marcus Hutter
2012 arXiv   pre-print
For the first time, without providing any domain knowledge, the same agent is able to self-adapt to a diverse range of interactive environments.  ...  For instance, AIXI is able to learn from scratch to play TicTacToe, Pacman, Kuhn Poker, and other games by trial and error, without even providing the rules of the games.  ...  What would be very interesting to show is that logic is an emergent phenomenon, i.e. that AIXI learns to reason logically if/since this helps collect reward.  ... 
arXiv:1202.6153v1 fatcat:y2pfo5s5ingm5dr3vtqj2ad4zi

AIs 10 to Watch: The Future of AI

2018 IEEE Intelligent Systems  
We don't know which factors govern a trained model's behavior (why it predicted what it did) and which kinds of patterns it can learn.  ...  When I say, "No thanks, I've had three already," you look back at the conversation to figure out what exactly I had three of. How can a computer do these things?  ... 
doi:10.1109/mis.2018.012001549 fatcat:mfx4dnb5lrgjbizilzh2anbcz4

Constructing interval variables via faceted Rasch measurement and multitask deep learning: a hate speech application [article]

Chris J. Kennedy, Geoff Bacon, Alexander Sahn, Claudia von Vacano
2020 arXiv   pre-print
The ratings on the theorized components of the target outcome are used as supervised, ordinal variables for the neural networks' internal concept learning.  ...  Our faceted Rasch scaling procedure integrates naturally with a multitask deep learning architecture for automated prediction on new data.  ...  For helpful comments and collaboration we are grateful to Rachel Rosen, Sebastian Raschka, Victor Vargas, Brittan Heller, D.  ... 
arXiv:2009.10277v1 fatcat:7qykkucus5hffdq5oghrkyoeqa

Artificial Intelligence [article]

John Paul Mueller Luca Massaron
2018 Zenodo  
So, the best way to start this book is to define what AI actually is, what it isn't, and how it connects to computers today  ...  Artificial Intelligence (AI) has had several false starts and stops over the years, partly because people don't really understand what AI is all about, or even what it should accomplish.  ...  You can divide machine learning algorithms into three main groups, based on their purpose: » Supervised learning » Unsupervised learning » Reinforcement learning The following sections discuss what different  ... 
doi:10.5281/zenodo.5599660 fatcat:rn6nyx5xinehzpwfy3knurtzgy

An Empirical Framework for Objective Testing for P-Consciousness in an Artificial Agent

Colin Hales
2009 Open Artificial Intelligence Journal  
The 'provability' derives from the delivery by science of objectively verifiable 'laws of nature'.  ...  Scientific behaviour is unique in being both highly formalised and provably critically dependent on the P-consciousness of the primary senses.  ...  David Grayden, Wanda Ginnane and unknown reviewers for helpful suggestions and comments.  ... 
doi:10.2174/1874061800903010001 fatcat:i7czzurhsfhhdo5vg3ovkk72qq
« Previous Showing results 1 — 15 out of 999 results