Data transformation pipelines shape the training distribution in computer vision, but designing them is a computationally expensive problem. We investigate whether large language models can enable efficient discovery by synthesizing executable transformation programs outside predefined augmentation spaces. We compare nine strategies under a unified training protocol, including standard augmentation policies, brute-force search, direct code generation with two LLMs, and performance-guided generation using LLMs fine-tuned on accuracy-ordered transformation pairs. When the best candidates are retrained for 20 epochs, LLM-generated transformation programs lead on every dataset, reaching 0.8877 accuracy on CIFAR-10, 0.5825 on CIFAR-100, and 0.9588 on SVHN, surpassing brute-force search over 6,000 candidates. The best transform on CIFAR-10 and SVHN comes from a fine-tuned generator using 270 candidates, while the best on CIFAR-100 comes from an off-the-shelf generator using only 10. This needs 22x fewer candidate evaluations in the generation step and 600x fewer for off-the-shelf generations, relative to brute-force search. Beyond transformation discovery, we analyze the reliability of the reduced-budget evaluation commonly used for augmentation search. Correlating one-epoch selection scores with 20-epoch outcomes reveals strong, weak, and absent predictive relationships depending on the dataset and architecture, with correlations ranging from 0.756 to near zero. A positive control indicates that range restriction alone is unlikely to explain the low correlations. Our results indicate that semantic synthesis with code LLMs provides an efficient alternative to search-based transformation discovery and highlight the importance of validating evaluation fidelity when optimizing data transformations.
Figures & tables
Figure 1: The fine-tuning loop. Each cycle samples reference transforms from the annotated pool, generates candidates (Listing ), evaluates them under the one-epoch protocol, admits those above a per-dataset threshold, and continues LoRA training on the rebuilt pair set. The loop runs 27 cycles.
Dataset
Architecture
N
n
Mean
Best
95% CI (mean)
CIFAR-10
ResNet
1
2000
0.5256
0.6124
[0.5234, 0.5279]
2
2000
0.4832
0.6071
[0.4801, 0.4863]
3
2000
0.4401
0.5983
[0.4363, 0.4439]
MobileNetV2
1
2000
0.3895
0.4604
[0.3877, 0.3914]
2
2000
0.3569
0.4655
[0.3545, 0.3593]
3
2000
0.3261
0.4517
[0.3231, 0.3290]
Table 1: Brute-force corpus by composition length, one-epoch evaluation. Candidates that failed to parse or execute are recorded at accuracy 0.0 and are included in n and in the mean.
Dataset
Strategy
Candidate
Acc@1ep
Acc@20ep
Δ rank
CIFAR-10
No augmentation
1
0.5832
0.7656
−3
AutoAugment
1
0.4083
0.7537
0
RandAugment
1
0.3554
0.7149
0
TrivialAugment
1
0.4673
0.8037
+1
Brute force
6000
0.6124
0.8115
−4
OlympicCoder-7B
10
0.4708
0.8728
+4
Table 2: Discovery strategies at both budgets, ResNet. Acc@1ep is the best one-epoch score over the strategy’s search budget; Acc@20ep is its best candidate retrained for 20 epochs. Δ rank is the change in within-dataset rank between budgets, positive meaning the strategy improves its position at full budget.
Table 3: How well the one-epoch selection score predicts the 20 -epoch outcome. Panel A correlates each retrained candidate’s own selection score against its own final accuracy, over the score span given. Panel B is a positive control: two independently trained one-epoch scores on CIFAR-10, over the whole corpus and then restricted to Panel A’s span. Coefficients are Spearman ρ with 95% Fisher z intervals.
Fully synthetic pretraining data generated by symbolic programs can bootstrap downstream training on naturalistic tasks, such as language modeling, but the generating programs have so far been designed by hand. We introduce PRECOCIAL, a method for automatically optimizing programs that generate synthetic pretraining data. Starting from an initial generator program, an LLM iteratively rewrites the program, and each rewrite is scored by pretraining a randomly-initialized transformer on its data and fine-tuning it on the downstream task. Because every score requires training, PRECOCIAL keeps a single network for the entire search, continuing its pretraining for each accepted program instead of training a new network per candidate, making it significantly more efficient. On four language and code datasets, at a fixed fine-tuning budget, optimized programs reach lower perplexity than hand-designed curricula, published procedural generators, random program mutation and the program the LLM writes in the first iteration of the search, and reach the same perplexity as the hand-designed curricula with 9-33% fewer fine-tuning steps. PRECOCIAL also ends with lower perplexity than FunSearch and AlphaEvolve, which train a new network for every candidate, at the same number of evaluations and a lower cost. We also observe that an optimized program generator from one domain transfers to datasets in other domains. Additionally, optimized programs allow for better downstream performance on four different reasoning and problem-solving tasks.
Large language models (LLMs) show strong potential for neural architecture generation, yet existing approaches produce complete model implementations from scratch -- computationally expensive and yielding verbose code. We propose Delta-Code Generation, where fine-tuned LLMs generate compact unified diffs (deltas) to refine baseline architectures rather than synthesizing entire models. Our pipeline iteratively fine-tunes the LLM via LoRA on curated architectures from the LEMUR dataset, with MinHash-Jaccard novelty filtering for structural diversity. We evaluate three 7B-class LLMs -- DeepSeek-Coder-7B, Qwen2.5-Coder-7B, and Mistral-7B -- across six datasets (CIFAR-10, CIFAR-100, MNIST, SVHN, ImageNette, CelebA) using a 22-cycle protocol (1,100 candidates per LLM). All three substantially surpass the full-generation baseline (50.6% valid rate, 42.3% mean first-epoch accuracy): DeepSeek-Coder reaches 75.3% valid rate and 65.8% mean accuracy; Qwen2.5-Coder 72.1%/64.6%; Mistral 66.6%/66.1%. On CIFAR-10, best first-epoch accuracies reach 85.5% (Mistral), 85.2% (DeepSeek), 80.6% (Qwen) -- well above 63.98% full generation and 71.5% for the concurrent approach of Gu et al. Output lengths are 30-50 lines versus 200+ for full generation (75-85% reduction). A 50-epoch study confirms the 1-epoch proxy preserves rankings (Mistral: Spearman ρ = 0.926). Delta-based generation is a token-efficient, multi-domain, LLM-agnostic alternative to full-model synthesis for LLM-driven NAS.
Santosh Premi Adhikari, Radu Timofte, Dmitry Ignatov
Computer Vision Lab, CAIDAS & IFI, University of Würzburg, Germany
Vision-language models (VLMs) still struggle with visual perception tasks such as spatial understanding and viewpoint recognition, largely because natural image datasets provide limited supervision for low-level visual skills. Can targeted synthetic supervision address these weaknesses without reference images or manual annotation? To investigate this, we introduce VisionFoundry, an automated pipeline that takes only a task name as input, uses LLMs to synthesize paired questions, answers, and text-to-image (T2I) prompts, generates images with T2I models, and filters samples via multimodal verification. With VisionFoundry, we construct VisionFoundry-10k, a synthetic VQA dataset spanning 10 perception tasks. Finetuning on VisionFoundry-10k consistently improves perception benchmarks across three open-source backbones (e.g., +6.7% on MMVP-pair and +10.5% on CV-Bench-3D for Qwen2.5-VL-3B-Instruct) while preserving broader capabilities and showing positive data scaling. The same synthetic supervision also yields consistent gains under reinforcement learning (RL) across all three backbones, and the framework remains effective under open-source synthesis and self-verification. Our findings demonstrate that automated synthetic supervision offers an effective and scalable path toward systematic VLM training.