Neural Information Processing Systems http://nips.cc/

We present a neural program synthesis approach integrating components which write, execute, and assess code to navigate the search space of possible programs. We equip the search process with an interpreter or a read-eval-print-loop (REPL), which immediately executes partially written programs, exposing their semantics. The REPL addresses a basic challenge of program synthesis: tiny changes in syntax can lead to huge changes in semantics. We train a pair of models, a policy that proposes the new piece of code to write, and a value function that assesses the prospects of the code written so-far. At test time we can combine these models with a Sequential Monte Carlo algorithm. We apply our approach to two domains: synthesizing text editing programs and inferring 2D and 3D graphics programs.

We train a pair of models, a policy that proposes the new piece of code to write, and a value function that assesses the prospects of the code written so-far.

We present linear prediction as a differentiable padding method. For each channel, a stochastic autoregressive linear model is fitted to the padding input by minimizing its noise terms in the least-squares sense. The padding is formed from the expected values of the autoregressive model given the known pixels. We trained the convolutional RVSR super-resolution model from scratch on satellite image data, using different padding methods. Linear prediction padding slightly reduced the mean square super-resolution error compared to zero and replication padding, with a moderate increase in time cost. Linear prediction padding better approximated satellite image data and RVSR feature map data. With zero padding, RVSR appeared to use more of its capacity to compensate for the high approximation error. Cropping the network output by a few pixels reduced the super-resolution error and the effect of the choice of padding method on the error, favoring output cropping with the faster replication and zero padding methods, for the studied workload.

"Given a large enough time budget the'no REPL' baseline is competitive with our ablated alternatives." However, the policy rollout baseline is trained with RL using a single machine, making it difficult to explore using entropy based methods or epsilon greedy. However, using multiple actors in an asynchronous setting would be a stronger/fairer baseline (and then doing policy rollouts) to the SMC approach. I expect SMC to do well but this is an important empirical question (other methods cited like Ganin et al. seem to do this in the same context). "The value-guided SMC sampler leads to the highest overall number of correct programs, requiring less time and fewer nodes expanded compared to other inference techniques. " - how well does a SMC sampler work without value guided proposals for both case studies?

This paper provides a method for Deep RL-based program synthesis, which exploits a SMC sampler during inference progressively decode into an executable program. The reviewers were enthusiastic about this method and found the experimental support for the proposal convincing. Without much need for further comment, I find the paper of acceptable standard for the conference. The authors are encouraged to take note of the suggestions made by the reviewers, especially R2 and R3, when improving the paper for eventual publication.

Large Language Models (LLMs) demonstrate strong abilities in common-sense reasoning and interactive decision-making, but often struggle with complex, long-horizon planning tasks. Recent techniques have sought to structure LLM outputs using control flow and other code-adjacent techniques to improve planning performance. These techniques include using variables (to track important information) and functions (to divide complex tasks into smaller re-usable sub-tasks). However, purely code-based approaches can be error-prone and insufficient for handling ambiguous or unstructured data. To address these challenges, we propose REPL-Plan, an LLM planning approach that is fully code-expressive (it can utilize all the benefits of code) while also being dynamic (it can flexibly adapt from errors and use the LLM for fuzzy situations). In REPL-Plan, an LLM solves tasks by interacting with a Read-Eval-Print Loop (REPL), which iteratively executes and evaluates code, similar to language shells or interactive code notebooks, allowing the model to flexibly correct errors and handle tasks dynamically. We demonstrate that REPL-Plan achieves strong results across various planning domains compared to previous methods.

We present a neural program synthesis approach integrating components which write, execute, and assess code to navigate the search space of possible programs. We equip the search process with an interpreter or a read-eval-print-loop (REPL), which immediately executes partially written programs, exposing their semantics. The REPL addresses a basic challenge of program synthesis: tiny changes in syntax can lead to huge changes in semantics. We train a pair of models, a policy that proposes the new piece of code to write, and a value function that assesses the prospects of the code written so-far. At test time we can combine these models with a Sequential Monte Carlo algorithm.

Hyperparameter tuning plays a crucial role in optimizing the performance of predictive learners. Cross--validation (CV) is a widely adopted technique for estimating the error of different hyperparameter settings. Repeated cross-validation (RCV) has been commonly employed to reduce the variability of CV errors. In this paper, we introduce a novel approach called blocked cross-validation (BCV), where the repetitions are blocked with respect to both CV partition and the random behavior of the learner. Theoretical analysis and empirical experiments demonstrate that BCV provides more precise error estimates compared to RCV, even with a significantly reduced number of runs. We present extensive examples using real--world data sets to showcase the effectiveness and efficiency of BCV in hyperparameter tuning. Our results indicate that BCV outperforms RCV in hyperparameter tuning, achieving greater precision with fewer computations.