{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/multi-task-learning-as-multi-objective","title":"Multi-Task Learning as Multi-Objective Optimization","arxiv_id":"1810.04650","date":"2018-10-10","proceeding":"NeurIPS 2018 12","authors":["Ozan Sener","Vladlen Koltun"],"abstract":"In multi-task learning, multiple tasks are solved jointly, sharing inductive\nbias between them. Multi-task learning is inherently a multi-objective problem\nbecause different tasks may conflict, necessitating a trade-off. A common\ncompromise is to optimize a proxy objective that minimizes a weighted linear\ncombination of per-task losses. However, this workaround is only valid when the\ntasks do not compete, which is rarely the case. In this paper, we explicitly\ncast multi-task learning as multi-objective optimization, with the overall\nobjective of finding a Pareto optimal solution. To this end, we use algorithms\ndeveloped in the gradient-based multi-objective optimization literature. These\nalgorithms are not directly applicable to large-scale learning problems since\nthey scale poorly with the dimensionality of the gradients and the number of\ntasks. We therefore propose an upper bound for the multi-objective loss and\nshow that it can be optimized efficiently. We further prove that optimizing\nthis upper bound yields a Pareto optimal solution under realistic assumptions.\nWe apply our method to a variety of multi-task deep learning problems including\ndigit classification, scene understanding (joint semantic segmentation,\ninstance segmentation, and depth estimation), and multi-label classification.\nOur method produces higher-performing models than recent multi-task learning\nformulations or per-task training.","url_abs":"http://arxiv.org/abs/1810.04650v2","url_pdf":"http://arxiv.org/pdf/1810.04650v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/IntelVCL/MultiObjectiveOptimization","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/VICO-UoE/KD4MTL","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/ebagdasa/backdoors101","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/isl-org/multiobjectiveoptimization","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/hav4ik/Hydra","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/salomonhotegni/mdmtn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"multi-task-learning-as-multi-objective","repo_url":"https://github.com/torchjd/torchjd","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"depth-estimation","task_name":"Depth Estimation"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"inductive-bias","task_name":"Inductive Bias"},{"task_slug":"instance-segmentation","task_name":"Instance Segmentation"},{"task_slug":"multi-label-classification-2","task_name":"MUlTI-LABEL-ClASSIFICATION"},{"task_slug":"multi-label-classification","task_name":"Multi-Label Classification"},{"task_slug":"multi-task-learning","task_name":"Multi-Task Learning"},{"task_slug":"scene-understanding","task_name":"Scene Understanding"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":null,"task_name":"valid"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/multi-task-learning-on-celeba","task":"Multi-Task Learning","dataset":"CelebA","model":"MGDA-UB","rank_in_archive_order":1,"of":1,"metrics":{"Error":"8.25"},"uses_additional_data":false},{"leaderboard":"/sota/multi-task-learning-on-cityscapes","task":"Multi-Task Learning","dataset":"Cityscapes test","model":"MultiObjectiveOptimization","rank_in_archive_order":3,"of":3,"metrics":{"mIoU":"66.63"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1810.04650","atlas_url":"https://app.syntology.ai/?focus=1810.04650","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1810.04650"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/isl-org/multiobjectiveoptimization","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/hav4ik/Hydra","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/salomonhotegni/mdmtn","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/ebagdasa/backdoors101","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/torchjd/torchjd","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/VICO-UoE/KD4MTL","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/IntelVCL/MultiObjectiveOptimization","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran_draft_wrong":2,"unverified":17},"by_repo_kind":{"official":{"samples":10,"ran":2,"repositories":1},"listed":{"samples":9,"ran":0,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"dbcb53696bc43ef9","entry":"conv3x3","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/models/pspnet.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/models/pspnet.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"dbcb53696bc43ef9"}},{"code_sha256_prefix":"fca7e85d958618dd","entry":"conv3x3_bn_relu","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/models/pspnet.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/models/pspnet.py","link_basis":"plan_row","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"fca7e85d958618dd"}},{"code_sha256_prefix":"0cf665cadafb5228","entry":"clusterization_solver","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/graph_clustering.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/graph_clustering.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"0cf665cadafb5228"}},{"code_sha256_prefix":"90e50bc6f1220bdd","entry":"conv3x3","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/models/resnet_mit.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/models/resnet_mit.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"90e50bc6f1220bdd"}},{"code_sha256_prefix":"99fd754b42a655dc","entry":"corrects","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/metrics.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/metrics.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"99fd754b42a655dc"}},{"code_sha256_prefix":"36497fd5247322ab","entry":"cross_entropy2d","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/losses.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/losses.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"36497fd5247322ab"}},{"code_sha256_prefix":"c500c5785882a8e8","entry":"get_losses","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/losses.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/losses.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c500c5785882a8e8"}},{"code_sha256_prefix":"7adcc62e674f9a40","entry":"get_metrics","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/metrics.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/metrics.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"7adcc62e674f9a40"}},{"code_sha256_prefix":"273e5eec3e68956d","entry":"get_metrics","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/metrics.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/metrics.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"273e5eec3e68956d"}},{"code_sha256_prefix":"faa96c0ba935fc57","entry":"gradient_normalizers","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/min_norm_solvers.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/min_norm_solvers.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"faa96c0ba935fc57"}},{"code_sha256_prefix":"e870ff8870d18569","entry":"l1_loss_depth","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/losses.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/losses.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"e870ff8870d18569"}},{"code_sha256_prefix":"a90eec2e662e141f","entry":"nll","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/losses.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/losses.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"a90eec2e662e141f"}},{"code_sha256_prefix":"66509d3a285284a7","entry":"normalize_grads","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/grad_normalizers.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/grad_normalizers.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"66509d3a285284a7"}},{"code_sha256_prefix":"3fc2b62ff7397e6f","entry":"prepare_dirs","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/log_utils.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/log_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"3fc2b62ff7397e6f"}},{"code_sha256_prefix":"aa26cdf2663e5a99","entry":"read_config","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/config_utils.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/config_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"aa26cdf2663e5a99"}},{"code_sha256_prefix":"deddf757c7ef17c5","entry":"resnet101","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/models/resnet_mit.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/models/resnet_mit.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"deddf757c7ef17c5"}},{"code_sha256_prefix":"db1b66fff0c3985f","entry":"resnet50","repo":"IntelVCL/MultiObjectiveOptimization","repo_kind":"official","path":"multi_task/models/resnet_mit.py","file_url":"https://github.com/IntelVCL/MultiObjectiveOptimization/blob/HEAD/multi_task/models/resnet_mit.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"db1b66fff0c3985f"}},{"code_sha256_prefix":"c3a6804df12ee19b","entry":"toy","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/datasets/toy.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/datasets/toy.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c3a6804df12ee19b"}},{"code_sha256_prefix":"f54c9b3ef3fd24f0","entry":"update_config","repo":"hav4ik/Hydra","repo_kind":"listed","path":"src/utils/config_utils.py","file_url":"https://github.com/hav4ik/Hydra/blob/HEAD/src/utils/config_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f54c9b3ef3fd24f0"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}