Algorithmic reasoning refers to the ability to understand the complex patterns behind the problem and decompose them into a sequence of reasoning steps towards the solution. Such nature of algorithmic reasoning makes it a challenge for large language models (LLMs), even though they have demonstrated promising performance in other reasoning tasks. Within this context, some recent studies use programming languages (e.g., Python) to express the necessary logic for solving a given instance/question (e.g., Program-of-Thought) as inspired by their strict and precise syntaxes. However, it is non-trivial to write an executable code that expresses the correct logic on the fly within a single inference call. Also, the code generated specifically for an instance cannot be reused for others, even if they are from the same task and might require identical logic to solve. This paper presents Think-and-Execute, a novel framework that decomposes the reasoning process of language models into two steps. (1) In Think, we discover a task-level logic that is shared across all instances for solving a given task and then express the logic with pseudocode; (2) In Execute, we further tailor the generated pseudocode to each instance and simulate the execution of the code. With extensive experiments on seven algorithmic reasoning tasks, we demonstrate the effectiveness of Think-and-Execute. Our approach better improves LMs' reasoning compared to several strong baselines performing instance-specific reasoning (e.g., CoT and PoT), suggesting the helpfulness of discovering task-level logic. Also, we show that compared to natural language, pseudocode can better guide the reasoning of LMs, even though they are trained to follow natural language instructions.
The exploration of transition state (TS) geometries is crucial for elucidating chemical reaction mechanisms and modeling their kinetics. In recent years, machine learning (ML) models have shown remarkable performance in TS geometry prediction. However, they require 3D geometries of reactants and products that can be challenging to determine. To tackle this, we introduce TSDiff, a novel ML model based on the stochastic diffusion method, which generates the 3D geometry of the TS from a 2D graph composed of molecular connectivity. Despite of this simple input, TSDiff generated TS geometries with high accuracy, outperforming existing ML models that utilize geometric information. Moreover, the generative model approach enabled the sampling of various valid TS conformations, even though only a single conformation for each reaction was used in training. Consequently, TSDiff also found more favorable reaction pathways with lower barrier heights than those in the reference database. We anticipate that this approach will be useful for exploring complex reactions that require the consideration of multiple TS conformations.
As quantum chemical properties have a significant dependence on their geometries, graph neural networks (GNNs) using 3D geometric information have achieved high prediction accuracy in many tasks. However, they often require 3D geometries obtained from high-level quantum mechanical calculations, which are practically infeasible, limiting their applicability in real-world problems. To tackle this, we propose a method to accurately predict the properties with relatively easy-to-obtain geometries (e.g., optimized geometries from the molecular force field). In this method, the input geometry, regarded as the corrupted geometry of the correct one, gradually approaches the correct one as it passes through the stacked denoising layers. We investigated the performance of the proposed method using 3D message-passing architectures for two prediction tasks: molecular properties and chemical reaction property. The reduction of positional errors through the denoising process contributed to performance improvement by increasing the mutual information between the correct and corrupted geometries. Moreover, our analysis of the correlation between denoising power and predictive accuracy demonstrates the effectiveness of the denoising process.
In this paper, we propose a novel benchmark called the StarCraft Multi-Agent Challenges+, where agents learn to perform multi-stage tasks and to use environmental factors without precise reward functions. The previous challenges (SMAC) recognized as a standard benchmark of Multi-Agent Reinforcement Learning are mainly concerned with ensuring that all agents cooperatively eliminate approaching adversaries only through fine manipulation with obvious reward functions. This challenge, on the other hand, is interested in the exploration capability of MARL algorithms to efficiently learn implicit multi-stage tasks and environmental factors as well as micro-control. This study covers both offensive and defensive scenarios. In the offensive scenarios, agents must learn to first find opponents and then eliminate them. The defensive scenarios require agents to use topographic features. For example, agents need to position themselves behind protective structures to make it harder for enemies to attack. We investigate MARL algorithms under SMAC+ and observe that recent approaches work well in similar settings to the previous challenges, but misbehave in offensive scenarios. Additionally, we observe that an enhanced exploration approach has a positive effect on performance but is not able to completely solve all scenarios. This study proposes new directions for future research.