Pith. sign in

REVIEW 5 major objections 5 minor 60 references

Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models

T0 review · 5 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read One small model can absorb two incompatible teachers and beat both on their own benchmarks.

desk verdict Solid new method for a real gap, with one load-bearing lemma parked in a missing appendix. read the letter →

arxiv 2608.04349 v1 pith:4STAFEGV submitted 2026-08-05 cs.CV

Siming Fu , Haojun Xu , Ruizhe He , Zheming Fu , Hualiang Wang , Jie Huang , Xiaoxiao Ma , Mingchen Zhong
show 4 more authors
Weihu Huang Xiaoxuan He Linjiang Huang Si Liu
This is my paper · ORCID
classification cs.CV
keywords multi-teacherdistillationon-policyflowmatchingtext-to-imagegenerationlatent-incompatiblemodelsLoRAadapterscompositionalgap-awarecurriculum
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Poly-OPD claims that a single small text-to-image model can inherit the complementary skills of two large teachers that cannot directly communicate with each other, because the teachers use different autoencoders and noise schedules, and can switch between those skills at inference time by swapping a small adapter. The paper's route is on-policy distillation through a pixel bridge: the student's own image is decoded to pixels, re-encoded in the chosen teacher's latent space, refined by that teacher from a matched noise level, and then compared to the student's own resumed rollout in a frozen DINOv2 feature space. Two further mechanisms make the consolidation stick: a gradient-compatibility measurement decides which parts of the network to share across capabilities, and a gap-aware curriculum spends more training on compositional categories where the student still lags the teacher. Reported on benchmarks, the distilled 2.5B student raises GenEval from 67.3 to 73.3, above both larger teachers, and lifts DrawBench HPSv3 from 9.34 to 11.35.

What carries the argument

The load-bearing object is the pixel bridge with noise-magnitude alignment (Eqs. 2–4). Since the student and each teacher share no latent coordinate system, the student's own generated image is decoded from student latents to pixels, re-encoded with the active teacher's encoder, re-noised to a noise level matched in magnitude to a student-schedule level, and refined by the frozen teacher for r remaining steps; the refined image is compared, through a frozen DINOv2 CLS embedding, against the student's reroll of its cached trajectory from the same noise magnitude. The same DINOv2 embedding is what makes supervision canonical across models. The second mechanism is the gradient-compatibility diagnostic, computed on paired samples that share noise and differ only in teacher, which determines that attention LoRA gradients are aligned across modes while FFN gradients are not, dictating that attention LoRA be shared and FFN adapters be per-capability. The third is gap-aware sampling, where category weights follow the smoothed positive residual between cached teacher scores and rolling student probe scores, annealing back toward uniform as gaps close.

What would settle it

Measure, at the end of training on a held-out prompt set, the DINOv2 cosine similarity between the teacher-refined target and the student's resumed rollout for the same prompt, seed, and noise magnitude; if the two are not consistently close, the aligned entry points that the distillation loss compares are semantically mismatched, and the pipeline's measured gains would not be attributable to on-policy correction.

Watch

Extended reading notes

Core claim

The central claim is that heterogeneous on-policy distillation is possible: a flow-matching student can learn from teachers whose latent coordinates are mutually unintelligible, provided the supervision is routed through pixels and measured in a semantic space. In the paper's formulation, the student sample is decoded and re-encoded into the active teacher's autoencoder, re-noised to a level matched by magnitude (not by timestep index), and refined by the frozen teacher; the refined image is then a fixed target in DINOv2 CLS space, and the student is trained to make its own resumed rollout match that target. The paper further claims that the sharing–isolation boundary between capabilities should be set by direct gradient measurement, yielding shared attention LoRA modules and teacher-specific feed-forward adapters, and that training budget within the compositional mode should follow the remaining teacher–student gap rather than raw difficulty. With this machinery, a 2.5B SD3.5-Medium student surpasses both a 12B teacher and a 6B teacher on GenEval and improves preference metrics on DrawBench, which the authors read as evidence that on-policy training can exceed the teacher's own sample distribution.

Load-bearing premise

The two-stage pipeline rests on the unstated lemma that once the student is near convergence, the teacher's re-noised, re-encoded version of the student's image and the student's own cached latent state at the matched noise level describe the same image content to first order; the proof is deferred to an appendix that is not present in the submitted text.

Editorial extensions

If this is right

  • A compact student can consolidate capabilities from multiple large, architecturally incompatible teachers, so serving several specialists no longer requires deploying several full models.
  • Capability selection at inference can cost an adapter swap rather than a second model's memory and latency.
  • The refinement-depth knob r interpolates from off-policy imitation (full teacher rerun) to local on-policy correction, giving a single continuous trade-off between how closely the student tracks the teacher and training cost.
  • Because the student can surpass its teachers on aggregated or category-level benchmarks, teacher distillation need not be bounded by the teacher's own sample distribution.
  • Gap-aware sampling, driven by the measured teacher–student residual, automatically anneals toward uniform sampling as capabilities saturate, removing the need for an external curriculum schedule.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The pixel-bridge scheme is not tied to diffusion or flow teachers: any generative model that can refine a decoded image, such as an autoregressive or GAN-based model, could in principle serve as an on-policy teacher for a flow-matching student as long as a semantic feature space like DINOv2 spans both.
  • The gradient-compatibility diagnostic could be reused as a general tool in multi-task and multi-teacher adaptation to decide, for any module type, whether to share parameters or isolate them, rather than fixing the boundary by architecture convention.
  • If the missing content-consistency lemma is confirmed, a testable prediction follows: the benefit of the on-policy stage over warm-start-only training should grow as the refinement depth r increases up to the point where the teacher fully regenerates the image, and should vanish if the teacher refinement is applied to semantically unrelated images.
  • The same framework invites a direct comparison against routing: a unified student with adapter switches versus a system that keeps both full teachers and routes prompts, where the measurable trade-off is benchmark quality per unit of memory and latency.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes Poly-OPD, a multi-teacher distillation framework for flow-matching text-to-image models whose teachers use different autoencoders and noise schedules. The method bridges the student and teacher latents through pixel space, applies DINOv2 CLS supervision in a common semantic space, uses a gradient-compatibility diagnostic to share attention LoRA while isolating per-teacher FFN adapters, and schedules compositional prompts by the remaining teacher-student gap. The main empirical claim is that distilling FLUX.1-dev and Z-Image into a 2.5B SD3.5-Medium student raises GenEval from 67.3 to 73.3 and DrawBench HPSv3 from 9.34 to 11.35, while enabling inference-time capability switching via adapter swaps. The paper includes ablations of warm start, adapter sharing, gap-aware sampling, refinement range, and perceptual feature space.

Significance. If the results hold, Poly-OPD is a meaningful step for multi-capability consolidation: it relaxes the shared-latent assumption that limits existing distillation methods, proposes a concrete pixel-bridge plus semantic-space supervision scheme, and grounds the adapter sharing boundary in a direct measurement rather than convention. The empirical study is unusually thorough in its ablations: the warm-start dependence (Table 4), the refinement-noise sweep (Table 5), and the feature-space comparison (Table 6) isolate the effect of each design choice, and the gradient-compatibility result (Figure 3) is a useful diagnostic in itself. The significance is conditional, however, because the correctness of the on-policy objective rests on an unproven content-consistency lemma, the headline numbers are reported without variance estimates, and the relationship between the composition training pool and the GenEval evaluation set is not made explicit.

major comments (5)
  1. [Sec. 3.2, Eqs. (3)-(4)] The claim that the teacher-refined target x_ref and the student re-rolled output x_tilde are 'content-consistent to first order once the student is near convergence' is load-bearing for the entire on-policy objective, because the DINOv2 loss only measures the teacher's correction if the two entry points represent the same semantic content. The manuscript refers to an Appendix for the proof, but no appendix is included in the submitted text. The warm-start ablation (Table 4, GenEval 49.2 vs 73.3) is consistent with the lemma but does not establish it; it is equally consistent with the loss being semantically meaningless when the student is far from the teacher. Please include the missing proof or, at minimum, an empirical check that reports the DINOv2 cosine similarity between x_ref and x_tilde at matched noise levels at the trained checkpoint and across refinement depths r.
  2. [Sec. 4, Tables 1-6] No error bars, confidence intervals, or multiple-seed results are reported for any of the main comparisons. The headline claims are point differences of roughly 6 GenEval points over the base student and 4 points over the Z-Image teacher, as well as HPSv3 differences of about 2 points; without variance estimates the reader cannot judge whether these differences are beyond run-to-run noise. Please provide at least three independent training runs (or bootstrap intervals over generated samples) for the full configuration and for the most important ablations, especially Tables 1, 2, and 4.
  3. [Sec. 4.1, Table 2] The composition-mode training pool is described as the 'GenEval-style split of Flow-GRPO' while evaluation is performed on GenEval itself. The probe set is stated to be disjoint from both the training pool and the evaluation set, but the manuscript does not state that the training prompts are disjoint from the 553 GenEval evaluation prompts or that they are not paraphrases of them. If the training pool overlaps with or is derived from the GenEval test prompts, the reported GenEval improvement would partly reflect memorization rather than compositional generalization. Please clarify the exact relationship between the Flow-GRPO split and the GenEval benchmark, and confirm that no GenEval evaluation prompt appears in the training pool or probe set.
  4. [Sec. 4.3, Table 6] The row 'Student latent MSE' reports 'Collapse' without a numeric value, and the text says the loss 'collapses outright.' This is a qualitative claim in a quantitative table, and it is used to justify the choice of DINOv2. Please quantify the collapse: report the metric values, training curves, or a clear failure mode (e.g., NaNs, divergence, or blank images) so the reader can verify that the comparison is not an artifact of a mis-specified MSE implementation.
  5. [Sec. 3.3, Eq. (6)] The gradient-compatibility diagnostic is computed at the warm-start initialization and with the Eq. (1) losses, and the resulting sharing-isolation boundary is then applied throughout the OPD stage. The manuscript does not establish that the gradient-compatibility pattern is stable as training progresses, even though the OPD objective differs from the warm-start objective. If the pattern shifts during training, the measured boundary may not be the optimal one. Please report the diagnostic at a later training point (or at least argue why the initial measurement is representative), since the adapter design is one of the central contributions.
minor comments (5)
  1. [Sec. 3.2, Eq. (5)] The symbol lambda_WS in Eq. (5) is used without a prior definition; please define it in the notation block or near the objective.
  2. [Fig. 2] The figure labels 'Student VAE Dec.' and 'Teacher VAE dec.' while the text in Eqs. (2)-(4) uses E_S, G_S, E_e, G_e; please align the figure notation with the equations for consistency.
  3. [Table 2] The GenEval 'Single' column for Poly-OPD is listed as 100.0 against a base of 99.06; please confirm this is not a typo and, if correct, mention it explicitly as a ceiling effect.
  4. [Throughout] The capitalization of 'Z-Image' is inconsistent across the text, tables, and references; please unify it.
  5. [Sec. 4.1] The evaluation protocol promises that 'Full sampler, guidance, and schedule settings are in the supplement,' but no supplement is included in the submitted text; if this is an artifact of the submission format, please ensure the supplement is supplied with the revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reduction: the distillation objective and headline benchmark gains are evaluated on external, held-out suites, and the only in-text caveat is a deferred proof, not a circular step.

full rationale

The paper's central claim (GenEval 67.3 to 73.3 and DrawBench HPSv3 9.34 to 11.35) is assessed on external benchmark suites (GenEval, DPG-Bench, DrawBench) with prompts disjoint from the training and probe pools, so the headline numbers are not fitted inputs renamed as predictions. The gradient-compatibility diagnostic (Sec. 3.3, Eq. 6) is a measurement that directly dictates the adapter sharing–isolation split, and the gap-aware sampler (Sec. 3.4, Eq. 8) uses held-out probes disjoint from evaluation; neither reduces to the target result by construction. The one passage that must be flagged explicitly is the deferred content-consistency proof for the on-policy entry points: 'Appendix shows that the two entry points into the aligned segment, namely the teacher re-noising the decoded image and the student resuming from its cached state, are content-consistent to first order once the student is near convergence' (Sec. 3.2, after Eq. 4). No such appendix appears in the provided manuscript text, so this is an omitted-support or verification gap for the on-policy objective. It is not a circular reduction, however: Eq. (3) and Eq. (4) are distinct computations from different latent codes and autoencoders, and the DINOv2 cosine loss is not fitted to the benchmarks. The citations to prior on-policy distillation works (Li et al. 2026; Fang et al. 2026) are used only to motivate the principle, not as the load-bearing argument, and no uniqueness theorem is imported from the authors' own prior work. The claim that surpassing the supervising teachers is 'evidence unique to on-policy training' is an interpretation of the experimental result, not a derivation from a fitted quantity. Overall, the derivation chain is self-contained against external benchmarks and contains no circularity.

Assumptions & free parameters 10 free parameters · 5 assumptions · 0 invented entities

The method is empirical and its central claim (benchmark improvements) is not derived from a fitted theory. The listed free parameters are training hyperparameters chosen by hand or ablation; they are not fitted to the evaluation benchmarks. The main axioms are the domain assumptions of the pixel bridge and DINOv2 semantic invariance, plus two ad hoc assumptions about content consistency and gradient-compatibility stability. No invented physical or conceptual entities are introduced.

free parameters (10)
  • lambda (mode prior) = not reported
    Sampling probability for the preference mode in Eq. (5); value is not given in the text.
  • beta (gap EMA momentum) = 0.8
    Hyperparameter in Eq. (8) that smooths the gap estimates, set in Sec 4.1.
  • tau (gap temperature) = 0.1
    Temperature controlling how greedily the largest gap is favored in Eq. (8), set in Sec 4.1.
  • k (probe prompts per category) = 32
    Number of held-out prompts per compositional category for gap estimation, set in Sec 4.1.
  • K (gap update interval) = 50
    Number of training steps between recomputing student probe scores, set in Sec 4.1.
  • Warm-start steps = 500
    Number of Stage 1 off-policy distillation steps, set in Sec 4.1.
  • OPD steps = 800
    Number of Stage 2 on-policy distillation steps, set in Sec 4.1.
  • Student Euler steps = 20
    Number of student rollout steps in Stage 2, set in Sec 4.1.
  • Refinement range [rmin, rmax] = [15, 20]
    Range of teacher refinement depths in Eq. (3) on the 20-step teacher grid, set in Sec 4.1 and ablated in Table 5.
  • lambda_WS (warm-start retention weight) = not reported
    Weight of the retained warm-start term in Eq. (5); value is not given in the text.
assumptions (5)
  • standard math Flow-matching interpolation and velocity objective (Eq. 1)
    The student objective relies on the standard flow-matching formulation of Lipman et al. and Liu et al., cited in Sec. 3.1.
  • domain assumption Pixels are the only shared coordinate system
    The pixel bridge (Eq. 2) assumes decoded pixels preserve enough content to re-encode into the teacher's latent; used in Sec. 3.2.
  • domain assumption DINOv2 CLS embeddings are semantically comparable across models
    The OPD loss (Eq. 4) uses cosine similarity in frozen DINOv2 space; the paper argues this is invariant to model-specific pixel statistics.
  • ad hoc to paper Content-consistency to first order between re-noised teacher latent and student cached latent
    Stated in Sec. 3.2, deferred to an appendix that is not present in the submitted text.
  • ad hoc to paper Gradient compatibility pattern is stable across training
    The adapter sharing/isolation decision (Sec. 3.3) is based on a measurement at warm-start initialization, assuming it persists during optimization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models." pith.science (2026). https://pith.science/paper/4STAFEGV

@misc{pith2026260804349,
  author       = {Pith},
  title        = {Pith review of: Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4STAFEGV}},
  note         = {Machine review of arXiv:2608.04349}
}
read the original abstract

Leading open text-to-image models often carry complementary strengths: one may lead on preference-aligned aesthetics while another follows compositional instructions more faithfully. However, differences in their autoencoders and noise schedules make it difficult to transfer these strengths across models. In this paper, we present Poly-OPD, a framework that can consolidate complementary strengths of heterogeneous teachers into a single compact flow-matching student. To bridge the incompatible latent spaces of different teachers, Poly-OPD performs on-policy distillation through a pixel bridge. Each student-generated image is re-encoded by a selected teacher's encoder and refined from a noise level matched by magnitude under the teacher's noise schedule. The resulting target is further matched to the student in frozen DINOv2 space, enabling supervision across incompatible latent spaces. To retain complementary capabilities without cross-teacher interference, Poly-OPD uses a gradient compatibility diagnostic to organize its adapters: attention LoRA modules are shared across teachers, whereas feed-forward adapters remain teacher-specific. During distillation, a gap-aware curriculum devotes more training to compositional categories where the student still falls short of the teacher. As each gap narrows, training shifts toward categories with larger remaining gaps. By distilling FLUX.1-dev and Z-Image into a 2.5B SD3.5-Medium student, Poly-OPD improves GenEval from 67.3 to 73.3, surpassing both larger teachers, and raises DrawBench HPSv3 from 9.34 to 11.35, consolidating both strengths within a switchable model.

Figures

Figures reproduced from arXiv: 2608.04349 by the authors.

Figure 1
Figure 1. Motivation and outcome of heterogeneous teacher distillation. Left: FLUX.1-dev produces visually appealing images but can miss prompt details such as spatial relations, colors, or object counts, whereas Z-Image better follows these compositional constraints but often has weaker visual appeal. Right: the radar chart compares FLUX.1-dev, Z-Image, the SD3.5-Medium base student, and Poly￾OPD, showing that the distilled … view at source ↗
Figure 2
Figure 2. Overview of Poly-OPD. The current student sample is bridged through pixels into the selected teacher latent space, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Gradient compatibility diagnostic. Bars re [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on representative prompts. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative effect of the warm-start stage. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 9 canonical work pages

  1. [1]

    arXiv preprint arXiv:2403.03206 , year=

    Scaling Rectified Flow Transformers for High-Resolution Image Synthesis , author=. arXiv preprint arXiv:2403.03206 , year=

    arXiv

  2. [2]

    2024 , howpublished=

    FLUX.1 , author=. 2024 , howpublished=

    2024

  3. [3]

    arXiv preprint arXiv:2202.00512 , year=

    Progressive Distillation for Fast Sampling of Diffusion Models , author=. arXiv preprint arXiv:2202.00512 , year=

    arXiv

  4. [4]

    International Conference on Learning Representations , volume=

    Improved techniques for training consistency models , author=. International Conference on Learning Representations , volume=

  5. [5]

    arXiv preprint arXiv:2311.18828 , year=

    One-step Diffusion with Distribution Matching Distillation , author=. arXiv preprint arXiv:2311.18828 , year=

    arXiv

  6. [6]

    arXiv preprint arXiv:2405.14867 , year=

    Improved Distribution Matching Distillation for Fast Image Synthesis , author=. arXiv preprint arXiv:2405.14867 , year=

    arXiv

  7. [7]

    arXiv preprint arXiv:2605.15055 , year=

    DiffusionOPD: A Unified Perspective of On-Policy Distillation in Diffusion Models , author=. arXiv preprint arXiv:2605.15055 , year=

    arXiv

  8. [8]

    arXiv preprint arXiv:2605.08063 , year=

    Flow-OPD: On-policy distillation for flow matching models , author=. arXiv preprint arXiv:2605.08063 , year=

    arXiv

Show all 60 references
  1. [9]

    Advances in neural information processing systems , volume=

    Flow-grpo: Training flow matching models via online rl , author=. Advances in neural information processing systems , volume=

  2. [10]

    International Conference on Learning Representations , volume=

    On-policy distillation of language models: Learning from self-generated mistakes , author=. International Conference on Learning Representations , volume=

  3. [11]

    Forty-first International Conference on Machine Learning , year=

    Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation , author=. Forty-first International Conference on Machine Learning , year=

  4. [12]

    ICLR , year=

    Flow Matching for Generative Modeling , author=. ICLR , year=

  5. [13]

    ICLR , year=

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow , author=. ICLR , year=

  6. [14]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Hpsv3: Towards wide-spectrum human preference score , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  7. [15]

    Advances in Neural Information Processing Systems , volume=

    Imagereward: Learning and evaluating human preferences for text-to-image generation , author=. Advances in Neural Information Processing Systems , volume=

  8. [16]

    ICLR , year=

    LoRA: Low-Rank Adaptation of Large Language Models , author=. ICLR , year=

  9. [17]

    arXiv preprint arXiv:2605.25378 , year=

    CollectionLoRA: Collecting 50 Effects in 1 LoRA via Multi-Teacher On-Policy Distillation , author=. arXiv preprint arXiv:2605.25378 , year=

    arXiv

  10. [18]

    Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining , pages=

    Learning from multiple teacher networks , author=. Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining , pages=

  11. [19]

    Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , pages=

    Mixture-of-loras: An efficient multitask tuning method for large language models , author=. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , pages=

    2024

  12. [20]

    arXiv preprint arXiv:2511.22699 , year=

    Z-image: An efficient image generation foundation model with single-stream diffusion transformer , author=. arXiv preprint arXiv:2511.22699 , year=

    arXiv

  13. [21]

    CVPR , year=

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric , author=. CVPR , year=

  14. [22]

    NeurIPS , year=

    Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding , author=. NeurIPS , year=

  15. [23]

    NeurIPS , year=

    Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation , author=. NeurIPS , year=

  16. [24]

    arXiv preprint arXiv:2403.05135 , year=

    Ella: Equip diffusion models with llm for enhanced semantic alignment , author=. arXiv preprint arXiv:2403.05135 , year=

    arXiv

  17. [25]

    Advances in Neural Information Processing Systems , volume=

    Geneval: An object-focused framework for evaluating text-to-image alignment , author=. Advances in Neural Information Processing Systems , volume=

  18. [26]

    arXiv preprint arXiv:2509.05952 , year=

    Coefficients-Preserving Sampling for Reinforcement Learning with Flow Matching , author=. arXiv preprint arXiv:2509.05952 , year=

  19. [27]

    arXiv preprint arXiv:2304.07193 , year=

    Dinov2: Learning robust visual features without supervision , author=. arXiv preprint arXiv:2304.07193 , year=

    arXiv

  20. [28]

    NeurIPS , year=

    Denoising Diffusion Probabilistic Models , author=. NeurIPS , year=

  21. [29]

    ICLR , year=

    Score-Based Generative Modeling through Stochastic Differential Equations , author=. ICLR , year=

  22. [30]

    arXiv preprint arXiv:2310.04378 , year=

    Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference , author=. arXiv preprint arXiv:2310.04378 , year=

    arXiv

  23. [31]

    arXiv preprint arXiv:2503.05236 , year=

    Unified reward model for multimodal understanding and generation , author=. arXiv preprint arXiv:2503.05236 , year=

    arXiv

  24. [32]

    2022 , howpublished =

    Christoph Schuhmann , title =. 2022 , howpublished =

    2022

  25. [33]

    International Conference on Learning Representations , volume=

    Minillm: Knowledge distillation of large language models , author=. International Conference on Learning Representations , volume=

  26. [34]

    arXiv preprint arXiv:2503.07067 , year=

    Distillm-2: A contrastive approach boosts the distillation of llms , author=. arXiv preprint arXiv:2503.07067 , year=

    arXiv

  27. [35]

    arXiv preprint arXiv:2602.12125 , year=

    Learning beyond teacher: Generalized on-policy distillation with reward extrapolation , author=. arXiv preprint arXiv:2602.12125 , year=

    arXiv

  28. [36]

    arXiv preprint arXiv:2603.07079 , year=

    Entropy-Aware On-Policy Distillation of Language Models , author=. arXiv preprint arXiv:2603.07079 , year=

    arXiv

  29. [37]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    A convnet for the 2020s , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  30. [38]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Sigmoid loss for language image pre-training , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  31. [39]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Amalgamating knowledge from heterogeneous graph neural networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  32. [40]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Amalgamating knowledge towards comprehensive classification , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  33. [41]

    International Conference on Machine Learning , pages=

    Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  34. [42]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Human preference score: Better aligning text-to-image models with human preference , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  35. [43]

    Advances in neural information processing systems , volume=

    Gradient surgery for multi-task learning , author=. Advances in neural information processing systems , volume=

  36. [44]

    arXiv preprint arXiv:2606.11155 , year=

    Mean Flow Distillation: Robust and Stable Distillation for Flow Matching Models , author=. arXiv preprint arXiv:2606.11155 , year=

    arXiv

  37. [45]

    arXiv preprint arXiv:2302.11289 , year=

    Recon: Reducing conflicting gradients from the root for multi-task learning , author=. arXiv preprint arXiv:2302.11289 , year=

    arXiv

  38. [46]

    arXiv preprint arXiv:2012.14913 , year=

    Transformer feed-forward layers are key-value memories , author=. arXiv preprint arXiv:2012.14913 , year=

    2012 arXiv

  39. [47]

    European Conference on Computer Vision , pages=

    Distilling diffusion models into conditional gans , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  40. [48]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Roma: Robust dense feature matching , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  41. [49]

    International Conference on Artificial Intelligence and Statistics , pages=

    Curriculum learning by optimizing learning dynamics , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2021 , organization=

    2021

  42. [50]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  43. [51]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  44. [52]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  45. [53]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  46. [54]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    1984 doi

  47. [55]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  48. [56]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  49. [57]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  50. [58]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  51. [59]

    2017 , eprint=

    Attention Is All You Need , author=. 2017 , eprint=

    2017

  52. [60]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.