{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:GVPEIQIUNHLB2MLJWFDFLDLFZU","short_pith_number":"pith:GVPEIQIU","canonical_record":{"source":{"id":"2601.02997","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-06T13:20:28Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"a3c69c7ac5065a43548230f9893cba3ede2b35f79b153e8510cf1396bbed3fb5","abstract_canon_sha256":"b97a2f26a6b09d16b43525191d0191d15711d0cb93b559532a159b8f836f913a"},"schema_version":"1.0"},"canonical_sha256":"355e44411469d61d3169b146558d65cd20bc8ce2e3f91a682fd1f30d716dbb07","source":{"kind":"arxiv","id":"2601.02997","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.02997","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"arxiv_version","alias_value":"2601.02997v2","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.02997","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_12","alias_value":"GVPEIQIUNHLB","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_16","alias_value":"GVPEIQIUNHLB2MLJ","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_8","alias_value":"GVPEIQIU","created_at":"2026-06-02T02:04:12Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:GVPEIQIUNHLB2MLJWFDFLDLFZU","target":"record","payload":{"canonical_record":{"source":{"id":"2601.02997","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-06T13:20:28Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"a3c69c7ac5065a43548230f9893cba3ede2b35f79b153e8510cf1396bbed3fb5","abstract_canon_sha256":"b97a2f26a6b09d16b43525191d0191d15711d0cb93b559532a159b8f836f913a"},"schema_version":"1.0"},"canonical_sha256":"355e44411469d61d3169b146558d65cd20bc8ce2e3f91a682fd1f30d716dbb07","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:12.913183Z","signature_b64":"HQJsb2mxJQKc5hDopWSqrHA/I+p/NUyWcYglvIRXL1KrWilVqqDQMSs5EhSONuRcAAiZ7FK3LfzXHUO3numPBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"355e44411469d61d3169b146558d65cd20bc8ce2e3f91a682fd1f30d716dbb07","last_reissued_at":"2026-06-02T02:04:12.912739Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:12.912739Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2601.02997","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"QiDdZBYVTYSHAX9kCifgXkfQEBOT2sGBIWeiqFdP8X20FpOqvCYXsnkLjrBX1nB3Puoac1KTeT1SwehVuEJaBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T16:39:00.831154Z"},"content_sha256":"14c06bef2f7de3799dbc6716e1be630a4b9bbf24bf7c7af21886c05f2f2966ae","schema_version":"1.0","event_id":"sha256:14c06bef2f7de3799dbc6716e1be630a4b9bbf24bf7c7af21886c05f2f2966ae"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:GVPEIQIUNHLB2MLJWFDFLDLFZU","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"From Memorization to Creativity: LLM as a Designer of Novel Neural Architectures","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures.","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Dmitry Ignatov, Radu Timofte, Waleed Khalid","submitted_at":"2026-01-06T13:20:28Z","abstract_excerpt":"Large language models (LLMs) excel in program synthesis, yet their capacity for neural architecture design -- balancing syntactic reliability, performance, and structural novelty -- remains underexplored. We present a closed-loop architecture synthesis pipeline within the NNGPT framework, in which a code-oriented LLM evolves over 22 supervised fine-tuning cycles. At each cycle, the LLM synthesizes PyTorch convolutional networks, validated via low-fidelity performance signals and filtered via a MinHash--Jaccard criterion to prevent structural redundancy before being incorporated into the LEMUR "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By grounding synthesis in execution feedback and novelty filtering, we demonstrate that iterative self-supervised fine-tuning reshapes an LLM into a task-specialized architectural prior -- improving generation reliability, proxy performance, and structural diversity.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Low-fidelity performance signals serve as reliable proxies for true model quality, and the MinHash-Jaccard filter removes redundancy without discarding useful novel structures that would perform well under full evaluation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Iterative LLM fine-tuning in a feedback loop with low-fidelity validation and MinHash-Jaccard novelty filtering shifts generation toward more valid, higher-accuracy, and structurally novel convolutional architectures on CIFAR-10 with cross-dataset transfer.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4a807d8498e5ff0bb920c69f54739053454ef3caf0befd730d187d566834ee13"},"source":{"id":"2601.02997","kind":"arxiv","version":2},"verdict":{"id":"478dec2d-2a4a-4cd6-af90-35dcd8c6069d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T17:32:39.000178Z","strongest_claim":"By grounding synthesis in execution feedback and novelty filtering, we demonstrate that iterative self-supervised fine-tuning reshapes an LLM into a task-specialized architectural prior -- improving generation reliability, proxy performance, and structural diversity.","one_line_summary":"Iterative LLM fine-tuning in a feedback loop with low-fidelity validation and MinHash-Jaccard novelty filtering shifts generation toward more valid, higher-accuracy, and structurally novel convolutional architectures on CIFAR-10 with cross-dataset transfer.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Low-fidelity performance signals serve as reliable proxies for true model quality, and the MinHash-Jaccard filter removes redundancy without discarding useful novel structures that would perform well under full evaluation.","pith_extraction_headline":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.02997/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"478dec2d-2a4a-4cd6-af90-35dcd8c6069d"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"eW8+5DW5Bhp5G6PcWJl9ED8xQ7+RzX/HgVxQzyzj+2sArRbLiUYI9SMH4WN6zHWzfPvwVJ3BFobo45nI0PwBAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T16:39:00.831730Z"},"content_sha256":"8a32a8795a152b81f774f44ab20c4775f4474f3a460fd3d6576a5bfc3d7cea6a","schema_version":"1.0","event_id":"sha256:8a32a8795a152b81f774f44ab20c4775f4474f3a460fd3d6576a5bfc3d7cea6a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/bundle.json","state_url":"https://pith.science/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-03T16:39:00Z","links":{"resolver":"https://pith.science/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU","bundle":"https://pith.science/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/bundle.json","state":"https://pith.science/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/state.json","well_known_bundle":"https://pith.science/.well-known/pith/GVPEIQIUNHLB2MLJWFDFLDLFZU/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:GVPEIQIUNHLB2MLJWFDFLDLFZU","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"b97a2f26a6b09d16b43525191d0191d15711d0cb93b559532a159b8f836f913a","cross_cats_sorted":["cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-06T13:20:28Z","title_canon_sha256":"a3c69c7ac5065a43548230f9893cba3ede2b35f79b153e8510cf1396bbed3fb5"},"schema_version":"1.0","source":{"id":"2601.02997","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.02997","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"arxiv_version","alias_value":"2601.02997v2","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.02997","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_12","alias_value":"GVPEIQIUNHLB","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_16","alias_value":"GVPEIQIUNHLB2MLJ","created_at":"2026-06-02T02:04:12Z"},{"alias_kind":"pith_short_8","alias_value":"GVPEIQIU","created_at":"2026-06-02T02:04:12Z"}],"graph_snapshots":[{"event_id":"sha256:8a32a8795a152b81f774f44ab20c4775f4474f3a460fd3d6576a5bfc3d7cea6a","target":"graph","created_at":"2026-06-02T02:04:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"By grounding synthesis in execution feedback and novelty filtering, we demonstrate that iterative self-supervised fine-tuning reshapes an LLM into a task-specialized architectural prior -- improving generation reliability, proxy performance, and structural diversity."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"Low-fidelity performance signals serve as reliable proxies for true model quality, and the MinHash-Jaccard filter removes redundancy without discarding useful novel structures that would perform well under full evaluation."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Iterative LLM fine-tuning in a feedback loop with low-fidelity validation and MinHash-Jaccard novelty filtering shifts generation toward more valid, higher-accuracy, and structurally novel convolutional architectures on CIFAR-10 with cross-dataset transfer."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures."}],"snapshot_sha256":"4a807d8498e5ff0bb920c69f54739053454ef3caf0befd730d187d566834ee13"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2601.02997/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) excel in program synthesis, yet their capacity for neural architecture design -- balancing syntactic reliability, performance, and structural novelty -- remains underexplored. We present a closed-loop architecture synthesis pipeline within the NNGPT framework, in which a code-oriented LLM evolves over 22 supervised fine-tuning cycles. At each cycle, the LLM synthesizes PyTorch convolutional networks, validated via low-fidelity performance signals and filtered via a MinHash--Jaccard criterion to prevent structural redundancy before being incorporated into the LEMUR ","authors_text":"Dmitry Ignatov, Radu Timofte, Waleed Khalid","cross_cats":["cs.CV"],"headline":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-06T13:20:28Z","title":"From Memorization to Creativity: LLM as a Designer of Novel Neural Architectures"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.02997","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-16T17:32:39.000178Z","id":"478dec2d-2a4a-4cd6-af90-35dcd8c6069d","model_set":{"reader":"grok-4.3"},"one_line_summary":"Iterative LLM fine-tuning in a feedback loop with low-fidelity validation and MinHash-Jaccard novelty filtering shifts generation toward more valid, higher-accuracy, and structurally novel convolutional architectures on CIFAR-10 with cross-dataset transfer.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Iterative fine-tuning with execution feedback and novelty filtering turns an LLM into a specialized generator of novel neural network architectures.","strongest_claim":"By grounding synthesis in execution feedback and novelty filtering, we demonstrate that iterative self-supervised fine-tuning reshapes an LLM into a task-specialized architectural prior -- improving generation reliability, proxy performance, and structural diversity.","weakest_assumption":"Low-fidelity performance signals serve as reliable proxies for true model quality, and the MinHash-Jaccard filter removes redundancy without discarding useful novel structures that would perform well under full evaluation."}},"verdict_id":"478dec2d-2a4a-4cd6-af90-35dcd8c6069d"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:14c06bef2f7de3799dbc6716e1be630a4b9bbf24bf7c7af21886c05f2f2966ae","target":"record","created_at":"2026-06-02T02:04:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"b97a2f26a6b09d16b43525191d0191d15711d0cb93b559532a159b8f836f913a","cross_cats_sorted":["cs.CV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-06T13:20:28Z","title_canon_sha256":"a3c69c7ac5065a43548230f9893cba3ede2b35f79b153e8510cf1396bbed3fb5"},"schema_version":"1.0","source":{"id":"2601.02997","kind":"arxiv","version":2}},"canonical_sha256":"355e44411469d61d3169b146558d65cd20bc8ce2e3f91a682fd1f30d716dbb07","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"355e44411469d61d3169b146558d65cd20bc8ce2e3f91a682fd1f30d716dbb07","first_computed_at":"2026-06-02T02:04:12.912739Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-02T02:04:12.912739Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HQJsb2mxJQKc5hDopWSqrHA/I+p/NUyWcYglvIRXL1KrWilVqqDQMSs5EhSONuRcAAiZ7FK3LfzXHUO3numPBg==","signature_status":"signed_v1","signed_at":"2026-06-02T02:04:12.913183Z","signed_message":"canonical_sha256_bytes"},"source_id":"2601.02997","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:14c06bef2f7de3799dbc6716e1be630a4b9bbf24bf7c7af21886c05f2f2966ae","sha256:8a32a8795a152b81f774f44ab20c4775f4474f3a460fd3d6576a5bfc3d7cea6a"],"state_sha256":"70b7acce643edca546a53b747982d0393dc27a63b924e75d5cf5d087777446e1"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"i4wo9N8Ul4Rgwq/CLddkrbwAS6G3uzChLW9kL5AQIO/+PVtN6NTGCdD+CPahlmZ09WV86Ko47ckplFu8FyqaDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-03T16:39:00.834769Z","bundle_sha256":"59775456a940001013919b3a26777f1a98ed26f2c6793ba742ec36ba2c700112"}}