Model-agnostic Measure of Generalization Difficulty

The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose w...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Boopathy, Akhilan, Liu, Kevin, Hwang, Jaedong, Ge, Shu, Mohammedsaleh, Asaad, Fiete, Ila
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Boopathy, Akhilan
Liu, Kevin
Hwang, Jaedong
Ge, Shu
Mohammedsaleh, Asaad
Fiete, Ila
description The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose what is to our knowledge the first model-agnostic measure of the inherent generalization difficulty of tasks. Our inductive bias complexity measure quantifies the total information required to generalize well on a task minus the information provided by the data. It does so by measuring the fractional volume occupied by hypotheses that generalize on a task given that they fit the training data. It scales exponentially with the intrinsic dimensionality of the space over which the model must generalize but only polynomially in resolution per dimension, showing that tasks which require generalizing over many dimensions are drastically more difficult than tasks involving more detail in fewer dimensions. Our measure can be applied to compute and compare supervised learning, reinforcement learning and meta-learning generalization difficulties against each other. We show that applied empirically, it formally quantifies intuitively expected trends, e.g. that in terms of required inductive bias, MNIST < CIFAR10 < Imagenet and fully observable Markov decision processes (MDPs) < partially observable MDPs. Further, we show that classification of complex images < few-shot meta-learning with simple images. Our measure provides a quantitative metric to guide the construction of more complex tasks requiring greater inductive bias, and thereby encourages the development of more sophisticated architectures and learning algorithms with more powerful generalization capabilities.
doi_str_mv 10.48550/arxiv.2305.01034
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2305_01034</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2305_01034</sourcerecordid><originalsourceid>FETCH-LOGICAL-a674-3dfe17e23233403f7ed7031eafc0c5194a133c5a89c3c1a0604d98246600580a3</originalsourceid><addsrcrecordid>eNotzrsKwjAUgOEsDqI-gJN9gdaTnqSXUbyDxcW9HNITCdRW0irq04uX6d9-PiGmEiKVaQ1z8g93j2IEHYEEVEOBRVtxHdK5abvemaBg6m6eg9YGW27YU-1e1Lu2CVbOWmdudf8ci4GluuPJvyNx2qxPy114OG73y8UhpCRVIVaWZcoxxogK0KZcpYCSyRowWuaKJKLRlOUGjSRIQFV5FqskAdAZEI7E7Lf9qsurdxfyz_KjL796fAP6CT4t</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Model-agnostic Measure of Generalization Difficulty</title><source>arXiv.org</source><creator>Boopathy, Akhilan ; Liu, Kevin ; Hwang, Jaedong ; Ge, Shu ; Mohammedsaleh, Asaad ; Fiete, Ila</creator><creatorcontrib>Boopathy, Akhilan ; Liu, Kevin ; Hwang, Jaedong ; Ge, Shu ; Mohammedsaleh, Asaad ; Fiete, Ila</creatorcontrib><description>The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose what is to our knowledge the first model-agnostic measure of the inherent generalization difficulty of tasks. Our inductive bias complexity measure quantifies the total information required to generalize well on a task minus the information provided by the data. It does so by measuring the fractional volume occupied by hypotheses that generalize on a task given that they fit the training data. It scales exponentially with the intrinsic dimensionality of the space over which the model must generalize but only polynomially in resolution per dimension, showing that tasks which require generalizing over many dimensions are drastically more difficult than tasks involving more detail in fewer dimensions. Our measure can be applied to compute and compare supervised learning, reinforcement learning and meta-learning generalization difficulties against each other. We show that applied empirically, it formally quantifies intuitively expected trends, e.g. that in terms of required inductive bias, MNIST &lt; CIFAR10 &lt; Imagenet and fully observable Markov decision processes (MDPs) &lt; partially observable MDPs. Further, we show that classification of complex images &lt; few-shot meta-learning with simple images. Our measure provides a quantitative metric to guide the construction of more complex tasks requiring greater inductive bias, and thereby encourages the development of more sophisticated architectures and learning algorithms with more powerful generalization capabilities.</description><identifier>DOI: 10.48550/arxiv.2305.01034</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Learning ; Statistics - Machine Learning</subject><creationdate>2023-05</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,778,883</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2305.01034$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2305.01034$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Boopathy, Akhilan</creatorcontrib><creatorcontrib>Liu, Kevin</creatorcontrib><creatorcontrib>Hwang, Jaedong</creatorcontrib><creatorcontrib>Ge, Shu</creatorcontrib><creatorcontrib>Mohammedsaleh, Asaad</creatorcontrib><creatorcontrib>Fiete, Ila</creatorcontrib><title>Model-agnostic Measure of Generalization Difficulty</title><description>The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose what is to our knowledge the first model-agnostic measure of the inherent generalization difficulty of tasks. Our inductive bias complexity measure quantifies the total information required to generalize well on a task minus the information provided by the data. It does so by measuring the fractional volume occupied by hypotheses that generalize on a task given that they fit the training data. It scales exponentially with the intrinsic dimensionality of the space over which the model must generalize but only polynomially in resolution per dimension, showing that tasks which require generalizing over many dimensions are drastically more difficult than tasks involving more detail in fewer dimensions. Our measure can be applied to compute and compare supervised learning, reinforcement learning and meta-learning generalization difficulties against each other. We show that applied empirically, it formally quantifies intuitively expected trends, e.g. that in terms of required inductive bias, MNIST &lt; CIFAR10 &lt; Imagenet and fully observable Markov decision processes (MDPs) &lt; partially observable MDPs. Further, we show that classification of complex images &lt; few-shot meta-learning with simple images. Our measure provides a quantitative metric to guide the construction of more complex tasks requiring greater inductive bias, and thereby encourages the development of more sophisticated architectures and learning algorithms with more powerful generalization capabilities.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Learning</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotzrsKwjAUgOEsDqI-gJN9gdaTnqSXUbyDxcW9HNITCdRW0irq04uX6d9-PiGmEiKVaQ1z8g93j2IEHYEEVEOBRVtxHdK5abvemaBg6m6eg9YGW27YU-1e1Lu2CVbOWmdudf8ci4GluuPJvyNx2qxPy114OG73y8UhpCRVIVaWZcoxxogK0KZcpYCSyRowWuaKJKLRlOUGjSRIQFV5FqskAdAZEI7E7Lf9qsurdxfyz_KjL796fAP6CT4t</recordid><startdate>20230501</startdate><enddate>20230501</enddate><creator>Boopathy, Akhilan</creator><creator>Liu, Kevin</creator><creator>Hwang, Jaedong</creator><creator>Ge, Shu</creator><creator>Mohammedsaleh, Asaad</creator><creator>Fiete, Ila</creator><scope>AKY</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20230501</creationdate><title>Model-agnostic Measure of Generalization Difficulty</title><author>Boopathy, Akhilan ; Liu, Kevin ; Hwang, Jaedong ; Ge, Shu ; Mohammedsaleh, Asaad ; Fiete, Ila</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a674-3dfe17e23233403f7ed7031eafc0c5194a133c5a89c3c1a0604d98246600580a3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Learning</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Boopathy, Akhilan</creatorcontrib><creatorcontrib>Liu, Kevin</creatorcontrib><creatorcontrib>Hwang, Jaedong</creatorcontrib><creatorcontrib>Ge, Shu</creatorcontrib><creatorcontrib>Mohammedsaleh, Asaad</creatorcontrib><creatorcontrib>Fiete, Ila</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Statistics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Boopathy, Akhilan</au><au>Liu, Kevin</au><au>Hwang, Jaedong</au><au>Ge, Shu</au><au>Mohammedsaleh, Asaad</au><au>Fiete, Ila</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Model-agnostic Measure of Generalization Difficulty</atitle><date>2023-05-01</date><risdate>2023</risdate><abstract>The measure of a machine learning algorithm is the difficulty of the tasks it can perform, and sufficiently difficult tasks are critical drivers of strong machine learning models. However, quantifying the generalization difficulty of machine learning benchmarks has remained challenging. We propose what is to our knowledge the first model-agnostic measure of the inherent generalization difficulty of tasks. Our inductive bias complexity measure quantifies the total information required to generalize well on a task minus the information provided by the data. It does so by measuring the fractional volume occupied by hypotheses that generalize on a task given that they fit the training data. It scales exponentially with the intrinsic dimensionality of the space over which the model must generalize but only polynomially in resolution per dimension, showing that tasks which require generalizing over many dimensions are drastically more difficult than tasks involving more detail in fewer dimensions. Our measure can be applied to compute and compare supervised learning, reinforcement learning and meta-learning generalization difficulties against each other. We show that applied empirically, it formally quantifies intuitively expected trends, e.g. that in terms of required inductive bias, MNIST &lt; CIFAR10 &lt; Imagenet and fully observable Markov decision processes (MDPs) &lt; partially observable MDPs. Further, we show that classification of complex images &lt; few-shot meta-learning with simple images. Our measure provides a quantitative metric to guide the construction of more complex tasks requiring greater inductive bias, and thereby encourages the development of more sophisticated architectures and learning algorithms with more powerful generalization capabilities.</abstract><doi>10.48550/arxiv.2305.01034</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2305.01034
ispartof
issn
language eng
recordid cdi_arxiv_primary_2305_01034
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Learning
Statistics - Machine Learning
title Model-agnostic Measure of Generalization Difficulty
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-15T14%3A21%3A01IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Model-agnostic%20Measure%20of%20Generalization%20Difficulty&rft.au=Boopathy,%20Akhilan&rft.date=2023-05-01&rft_id=info:doi/10.48550/arxiv.2305.01034&rft_dat=%3Carxiv_GOX%3E2305_01034%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true