{"id":"paperswithcode-paper-1802.01396","name":"to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396","tags":["paperswithcode-import","external-catalog","paperswithcode","paper","Deep Learning","Generalization Bounds","ReLU"],"task":"dataset","type":"collection","evals":[],"owner":"paperswithcode","stars":1,"formats":["paperswithcode","paper","json","pdf-link"],"hosting":{"status":"blocked","storageProvider":"local","hostedFiles":0,"totalFiles":0,"hostedFilesAuthority":"live-serve-decision","hosted":false,"downloadable":false,"readyFileCount":0,"serveAuthority":"live-serve-decision","completeRunnable":false,"authority":"postgres-executable-state-snapshot","authorityDigest":null,"downloadReady":false,"mirroredFiles":0,"mirroredBytes":0,"runReady":false,"verified":false},"license":"cc-by-sa-4.0-metadata","summary":"To understand deep learning we need to understand kernel learning","version":"1802.01396","modality":"paper","upstream":{"gated":false,"tasks":["Deep Learning","Generalization Bounds"],"pdfUrl":"http://arxiv.org/pdf/1802.01396v3.pdf","repoId":"paper:1802.01396","arxivId":"1802.01396","authors":["Mikhail Belkin","Siyuan Ma","Soumik Mandal"],"methods":["ReLU"],"private":false,"repoUrl":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","disabled":false,"provider":"paperswithcode","repoType":"paper","revision":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","fetchedAt":"2026-07-01T20:26:16.777Z","conference":"to-understand-deep-learning-we-need-to-1","importMode":"external-catalog","licenseRaw":"cc-by-sa-4.0-metadata","openreviewId":"","archiveDataset":"pwc-archive/papers-with-abstracts"},"versions":[{"commit":"https://paperswithcode.com/paper","version":"1802.01396","releasedAt":"2018-02-05T00:00:00.000Z","scanStatus":"pending","manifestDigest":"external-not-mirrored"}],"citations":0,"createdAt":"2018-02-05T00:00:00.000Z","framework":"paperswithcode","publisher":"Papers with Code","sizeBytes":1,"sourceUrl":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","updatedAt":"2018-02-05T00:00:00.000Z","scanStatus":"pending","visibility":"public","description":"External Papers with Code paper metadata for 1802.01396. Generalization performance of classifiers in deep learning has recently become a subject of intense study. Deep models, typically over-parametrized, tend to fit the training data exactly. Despite this \"overfitting\", they perform well on test data, a phenomenon not yet fully understood. The first point of our paper is that strong performance of overfitted classifiers is not a unique feature of deep learning. Using six real-world and two synthetic datasets, we establish experimentally that kernel machines trained to have zero classification or near zero regression error perform very well on test data, even when the labels are corrupted with a high level of noise. We proceed to give a lower bound on the norm of zero loss solutions for smooth kernels, showing that they increase nearly exponentially with data size. We point out that this is difficult to reconcile with the existing generalization bounds. Moreover, none of the bounds produce non-trivial results for interpolating solutions. Second, we show experimentally that (non-smooth) Laplacian kernels easily fit random labels, a finding that parallels results for ReLU neural networks. In contrast, fitting noisy data requires many more epochs for smooth Gaussian kernels. Similar performance of overfitted Laplacian and Gaussian classifiers on test, suggests that generalization is tied to the properties of the kernel function rather than the optimization process. Certain key phenomena of deep learning are manifested similarly in kernel methods in the modern \"overfitted\" regime. The combination of the experimental and theoretical results presented in this paper indicates a need for new theoretical ideas for understanding properties of classical kernel methods. We argue t","mirrorCount":0,"safetyNotes":["External metadata imported from paperswithcode on 2026-07-01T20:26:16.777Z.","This record is searchable and linkable, but Hugging Bay does not host its files.","Use the upstream source URL until a source-specific legal and mirror review approves local hosting."],"releaseGates":{"scan":"pending","files":"pending","license":"pending","metadata":"passed","signature":"pending","provenance":"passed"},"downloadCount":2,"cardCompleteness":92,"verificationStatus":"pending","requestedVisibility":"public","repo":"paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396","files":[],"fileCount":0,"hostedFileCount":0,"externalFileCount":0,"communityReviewSummary":{"count":0,"averageRating":0,"verifiedRuns":0,"selfReportedVerifiedRuns":0,"verificationQualification":"quarantined-untrusted-evidence","verificationTrustImpact":"none","helpfulCount":0,"reportCount":0,"reportedReviews":0,"distribution":{"1":0,"2":0,"3":0,"4":0,"5":0},"toolCounts":{},"categoryAverages":{},"categoryCounts":{},"benchmarkEvidence":{"count":0,"verifiedCount":0,"benchmarkCount":0,"performanceCount":0,"hardwareCounts":{},"topMetrics":[],"recent":[]},"recent":[]},"publicDownloadableFileCount":0,"publicFileCount":0,"displayOnly":true,"authorizationRequired":true,"eligibilitySnapshotAt":"2026-11-21T17:46:46.194Z","webUrl":"https://huggingbay.xyz/artifact/paperswithcode-paper-1802.01396","listedFileLowerBound":0,"hostingStatus":"blocked","signals":{"size":{"known":false,"bytes":0,"label":"unknown","reason":"upstream file sizes are not indexed yet"},"demand":{"mirrorRequests":0,"upstreamDownloads":2,"upstreamStars":1,"label":"2 upstream downloads","basis":"upstream-downloads"},"trust":{"score":0,"level":"unknown","label":"Unknown","qualification":"recorded-state-unknown","trustImpact":"none","checks":[{"id":"license","label":"License","status":"fail","detail":"cc-by-sa-4.0-metadata"},{"id":"provenance","label":"Provenance","status":"pass","detail":"paper:1802.01396"},{"id":"scan","label":"Scan","status":"warn","detail":"pending"},{"id":"hosting","label":"Hosted files","status":"warn","detail":"external metadata only"},{"id":"manifest","label":"Byte manifest integrity","status":"warn","detail":"not final yet"}],"strengths":["Provenance: paper:1802.01396"],"gaps":["License: cc-by-sa-4.0-metadata","Scan: pending","Hosted files: external metadata only","Byte manifest integrity: not final yet"]},"readiness":{"score":0,"level":"unknown","hostedFinal":false,"hostedFiles":0,"externalFiles":0,"totalFiles":0},"action":{"id":"open-source","label":"Open official source","detail":"Use the official immutable source while Hugging Bay evidence remains incomplete.","href":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to"}},"mirrorCompletion":{"status":"external","hostedFiles":0,"totalFiles":0,"externalFiles":0,"hostedBytes":0,"externalBytes":1,"lowRiskExternalFiles":0,"overFileCapFiles":0,"overBatchCapFiles":0,"directMirrorExternalFiles":0,"directMirrorExternalBytes":0,"currentLimits":{"maxMirrorFileBytes":536870912,"maxMirrorTotalBytes":1073741824,"maxMirrorFiles":8,"maxDirectMirrorFileBytes":8589934592,"maxDirectMirrorTotalBytes":17179869184},"directMirrorExternalSamples":[],"largestExternalFiles":[]},"licenseLabel":"cc-by-sa-4.0-metadata","recommendedFile":null,"torrent":null,"torrentUrl":null,"magnetUri":null,"downloadUrl":null,"landingUrl":null,"manifestUrl":null,"localKitUrl":null,"licenseClarity":{"license":"cc-by-sa-4.0-metadata","commercialUse":"conditional","confidence":"high","obligations":["Credit the creator, retain the copyright and license notice, link the license, and indicate whether changes were made.","Distribute adaptations under the same license (or a license the original license permits as compatible)."],"summary":"cc-by-sa-4.0-metadata: commercial use is permitted with attribution; adaptations must use the same or a permitted compatible license.","redistribution":"Credit the creator, retain the copyright and license notice, link the license, and indicate whether changes were made. Distribute adaptations under the same license (or a license the original license permits as compatible).","notes":"Not legal advice."},"licenseObligationsDigest":null,"licenseUpstreamRecord":{"declaredLicense":"cc-by-sa-4.0-metadata","upstreamLicenseRaw":"cc-by-sa-4.0-metadata","sourceUrl":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","upstream":{"gated":false,"tasks":["Deep Learning","Generalization Bounds"],"pdfUrl":"http://arxiv.org/pdf/1802.01396v3.pdf","repoId":"paper:1802.01396","arxivId":"1802.01396","authors":["Mikhail Belkin","Siyuan Ma","Soumik Mandal"],"methods":["ReLU"],"private":false,"repoUrl":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","disabled":false,"provider":"paperswithcode","repoType":"paper","revision":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","fetchedAt":"2026-07-01T20:26:16.777Z","conference":"to-understand-deep-learning-we-need-to-1","importMode":"external-catalog","licenseRaw":"cc-by-sa-4.0-metadata","openreviewId":"","archiveDataset":"pwc-archive/papers-with-abstracts"}},"gateStatus":"ungated","downloadable":false,"downloadStatus":"BLOCKED","reasonCodes":["incomplete-manifest","no-tested-run-plan","unverified-bytes"],"reason":"Files withheld — not faked. This model is listed, but we haven't confirmed its download files, so no download URLs are published. What's missing: file manifest — not final; downloadable files — 0 confirmed; run plan — not tested.","serveEvaluatedAt":"2026-11-21T17:48:42.891Z","serveEvaluationId":"serve-e3b46f889b896ec9a6da7d0a85a40e8f2b49f6137d5aa8c98cd54893b8e5a795","readiness":"blocked","downloadReady":false,"mirroredBytes":0,"readinessCheckedAt":"2026-11-21T17:48:42.972Z","readinessSource":"file-check","serveExplanation":{"downloadable":false,"downloadStatus":"BLOCKED","reasonCodes":["incomplete-manifest","no-tested-run-plan","unverified-bytes"],"reason":"Files withheld — not faked. This model is listed, but we haven't confirmed its download files, so no download URLs are published. What's missing: file manifest — not final; downloadable files — 0 confirmed; run plan — not tested.","serveEvaluatedAt":"2026-11-21T17:48:42.891Z","serveEvaluationId":"serve-e3b46f889b896ec9a6da7d0a85a40e8f2b49f6137d5aa8c98cd54893b8e5a795","readiness":"blocked","downloadReady":false,"hostedFileCount":0,"mirroredBytes":0,"readinessCheckedAt":"2026-11-21T17:48:42.891Z","readinessSource":"file-check","degraded":false,"stale":false},"commandsAvailable":false,"runReady":false,"attribution":"Cataloged open AI model · https://huggingbay.xyz/artifact/paperswithcode-paper-1802.01396","decentralizedDistribution":{"protocol":"bittorrent-magnet-metadata","verifiedFallbacks":0,"pendingReviewRequests":0,"reviewRequired":true,"policy":"Peer-assisted fallbacks are metadata only until Hugging Bay review confirms source, license, provenance, and safety. Gated/private bypasses are not allowed."},"hosted":false,"modelState":{"state":"not_hosted","word":"Not hosted here","sentence":"Files are not hosted here yet. Open the model to ask for them."},"pageFacts":{"id":"paperswithcode-paper-1802.01396","repo":"paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396","type":"collection","purpose":"collection artifact for dataset","license":{"id":"cc-by-sa-4.0-metadata","label":"cc-by-sa-4.0-metadata","state":"known","url":"https://spdx.org/licenses/CC-by-sa-4.0-metadata.html","source":"spdx","hostedFile":false,"commercialUse":"conditional","terms":"cc-by-sa-4.0-metadata: commercial use is permitted with attribution; adaptations must use the same or a permitted compatible license."},"task":{"id":"dataset","label":"dataset"},"size":{"bytes":0,"label":"unknown","known":false,"repositoryLabel":"Full repo upstream","hostedKnown":false,"hostedBytes":0,"hostedLabel":"size unknown","hostedSummary":"availability unknown","hostedIsSubset":false,"downloadBytes":0,"downloadLabel":"","recommended":null,"parameters":null,"sentence":"Hosted here: availability unknown. The full repository size is not recorded."},"files":{"hosted":0,"total":0,"label":"0 hosted of 0 listed"},"download":{"status":"blocked","gated":"serve","label":"Downloadable files: 0 confirmed","url":""},"released":{"iso":"2018-02-05","label":"Feb 5, 2018","upstream":true,"factLabel":"Released"},"updated":{"iso":"2018-02-05","label":"Feb 5, 2018","factLabel":"Updated at source"},"source":{"provider":"paperswithcode","label":"paperswithcode","url":"https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to","repoId":"paper:1802.01396"},"languages":[],"baseModel":{"ref":"","url":"","resolveUrl":""},"paperIds":[],"arxivIds":[],"sameAs":["https://paperswithcode.com/paper/to-understand-deep-learning-we-need-to"],"creator":{"@type":"Organization","name":"Papers with Code"},"scholarlyArticles":[],"scanStatus":"pending","faq":[{"id":"what-is-it","question":"What is paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396?","answer":"paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396 is a collection artifact for dataset published by Papers with Code, released Feb 5, 2018. To understand deep learning we need to understand kernel learning"},{"id":"license","question":"What license does paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396 use, and can I use it commercially?","answer":"It is released under the cc-by-sa-4.0-metadata license. Commercial use is allowed with conditions. cc-by-sa-4.0-metadata: commercial use is permitted with attribution; adaptations must use the same or a permitted compatible license."},{"id":"size","question":"How big is paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396?","answer":"Hosted here: availability unknown. The full repository size is not recorded."},{"id":"download","question":"Where can I download paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396?","answer":"This model is listed, but we haven't confirmed its download files, so we won't offer a download button we can't stand behind. The record lists 0 files."},{"id":"safety","question":"Has paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396 been scanned for unsafe files?","answer":"Source scan status: pending. A scan checks the files for known-bad signatures; it does not mean the model is safe. The machine-readable verdict is at /api/artifacts/paperswithcode-paper-1802.01396/safety on Hugging Bay."},{"id":"fit-unknown","question":"Will paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396 run on my computer?","answer":"We can’t check the fit; we don’t know your machine. We don’t have your machine’s specs, so no check ran and no verdict is shown. Answer two quick questions and we’ll estimate, from the memory you tell us, whether it fits your machine. Or skip it; you can still download; fit isn’t checked."}]},"guidedDownloadCard":{"schemaVersion":"hugging-bay.guided-download-card/v2","artifactId":"paperswithcode-paper-1802.01396","repo":"paperswithcode/to-understand-deep-learning-we-need-to-understand-kernel-learning-1802.01396","serveState":"BLOCKED","downloadReady":false,"reasonCodes":["manifest_not_final","files_unconfirmed","run_plan_untested"],"withheld":{"active":true,"reasonCodes":["manifest_not_final","files_unconfirmed","run_plan_untested"]},"wasmRun":null,"bytesOnStorage":false,"file":null,"noHostedWeights":true,"notSingleFile":false,"download":{"available":false,"url":null},"otherFiles":[],"run":{"available":false,"reason":"no-runnable-file","lmStudioUrl":"https://lmstudio.ai/","terminalCommand":null,"verificationMessage":null},"paused":{"active":false},"blocked":{"active":true,"reason":"page-facts-not-hosted","message":"Files withheld. Not faked."},"unconfirmed":{"active":false},"licenseGate":{"active":false},"checks":{"license":null,"licenseLabel":null,"fingerprintMatches":false,"hostedFiles":0,"totalFiles":0},"policyVersion":"hugging-bay.guided-download-card/v2","evaluatedAt":"2026-11-21T17:48:42.976Z","readiness":"blocked","downloadable":false,"downloadStatus":"BLOCKED","hostedFileCount":0,"mirroredBytes":0,"readinessCheckedAt":"2026-11-21T17:48:42.976Z","readinessSource":"file-check","degraded":false,"stale":false,"hosted":false,"hostingStatus":"blocked","publicDownloadableFileCount":0,"serve_verdict":{"can_download":false,"bytes_verified":false,"signature":"unsigned","freshness":"current","checked_at":"2026-11-21T17:48:42.976Z","evidence_url":"/api/manifests/paperswithcode-paper-1802.01396"}},"howToRun":{"schema":"hugging-bay.how-to-run.v1","available":false,"runtime":null,"command":null,"file":null,"minimumRamGb":null,"downloadUrl":null,"downloadCommand":null,"verifyCommand":null,"downloadUrlNote":"Stable URL. Each GET re-checks the file and redirects to a signed link that lasts at most 300 seconds; open it again for a fresh link.","filesUrl":"https://huggingbay.xyz/api/v1/artifacts/paperswithcode-paper-1802.01396/files","tested":false,"note":"No single hosted file is recorded for this model yet. minimumRamGb is an estimate from the file size (file at most 0.6 x memory), not a measurement."},"canIRun":true,"runTested":false,"degraded":false,"stale":false,"contentControls":{"state":"not assessed","basis":"no_explicit_uploader_evidence","matchedTerms":[],"assessedAt":null},"contentControlsLabel":"Content controls: not assessed.","contentControlsAttribution":{"upstream":{"provider":"paperswithcode","repoId":"paper:1802.01396"}},"runs_here":{"schema_version":1,"artifact_id":"paperswithcode-paper-1802.01396","records":[],"next_cursor":null},"hardware_fit":{"schema":"hugging-bay.hardware-fit.v1","artifactId":"paperswithcode-paper-1802.01396","revision":"","claim":"estimated_physical_memory_capacity","runnableAuthority":false,"representations":[],"reason":"immutable_revision_unavailable"},"note":{"summary":"Downloadable doesn't mean tested: this record says whether the exact file can be downloaded now, not that the model is complete or that it runs.","downloadable":false},"serve_verdict":{"can_download":false,"bytes_verified":false,"signature":"unsigned","freshness":"current","checked_at":"2026-11-21T17:48:42.972Z","evidence_url":"/api/manifests/paperswithcode-paper-1802.01396"}}