Large Language Models (LLMs) are increasingly utilized in scientific research assessment, particularly in automated paper review. However, existing LLMbased review systems face significant challenges, including limited domain expertise, hallucinated reasoning, and a lack of structured evaluation. To address these limitations, we introduce DeepReview, a multi-stage framework designed to emulate expert reviewers by incorporating structured analysis, literature retrieval, and evidence-based argumentation. Using DeepReview-13K, a curated dataset with structured annotations, we train DeepReviewer-14B, which outperforms CycleReviewer-70B with fewer tokens. In its best mode, DeepReviewer-14B achieves win rates of 88.21% and 80.20% against GPT-o1 and DeepSeek-R1 in evaluations. Our work sets a new benchmark for LLM-based paper review, with all resources publicly available. The code, model, dataset and demo have be released in http://ai-researcher.net.

We present an ngram model-based logit scaling technique that effectively transfers extreme subword stylistic variation to large language models at inference time. We demonstrate its efficacy by tracking the perplexity of generated text with respect to the ngram interpolated and original versions of an evaluation model. Minimizing the former measure while the latter approaches the perplexity of a text produced by a target author or character lets us select a sufficient degree of adaptation while retaining fluency.

Reading relevant scientific papers and analyzing research development trends is a critical step in generating new scientific ideas. However, the rapid increase in the volume of research literature and the complex citation relationships make it difficult for researchers to quickly analyze and derive meaningful research trends. The development of large language models (LLMs) has provided a novel approach for automatically summarizing papers and generating innovative research ideas. However, existing paper-based idea generation methods either simply input papers into LLMs via prompts or form logical chains of creative development based on citation relationships, without fully exploiting the semantic information embedded in these citations. Inspired by knowledge graphs and human cognitive processes, we propose a framework called the Graph of AI Ideas (GoAI) for the AI research field, which is dominated by open-access papers. This framework organizes relevant literature into entities within a knowledge graph and summarizes the semantic information contained in citations into relations within the graph. This organization effectively reflects the relationships between two academic papers and the advancement of the AI research field. Such organization aids LLMs in capturing the current progress of research, thereby enhancing their creativity. Experimental results demonstrate the effectiveness of our approach in generating novel, clear, and effective research ideas.

Evaluating Large Language Models (LLMs) free-form generated responses remains a challenge due to their diverse and open-ended nature. Traditional supervised signal-based automatic metrics fail to capture semantic equivalence or handle the variability of open-ended responses, while human evaluation, though reliable, is resource-intensive. Leveraging LLMs as evaluators offers a promising alternative due to their strong language understanding and instruction-following capabilities. Taking advantage of these capabilities, we propose the Dynamic Arbitration Framework for Evaluation (DAFE), which employs two primary LLM-as-judges and engages a third arbitrator only in cases of disagreements. This selective arbitration prioritizes evaluation reliability while reducing unnecessary computational demands compared to conventional majority voting. DAFE utilizes task-specific reference answers with dynamic arbitration to enhance judgment accuracy, resulting in significant improvements in evaluation metrics such as Macro F1 and Cohen's Kappa. Through experiments, including a comprehensive human evaluation, we demonstrate DAFE's ability to provide consistent, scalable, and resource-efficient assessments, establishing it as a robust framework for evaluating free-form model outputs.

Multimodal Audio-Language Models (ALMs) can understand and reason over both audio and text. Typically, reasoning performance correlates with model size, with the best results achieved by models exceeding 8 billion parameters. However, no prior work has explored enabling small audio-language models to perform reasoning tasks, despite the potential applications for edge devices. To address this gap, we introduce Mellow, a small Audio-Language Model specifically designed for reasoning. Mellow achieves state-of-the-art performance among existing small audio-language models and surpasses several larger models in reasoning capabilities. For instance, Mellow scores 52.11 on MMAU, comparable to SoTA Qwen2 Audio (which scores 52.5) while using 50 times fewer parameters and being trained on 60 times less data (audio hrs). To train Mellow, we introduce ReasonAQA, a dataset designed to enhance audio-grounded reasoning in models. It consists of a mixture of existing datasets (30% of the data) and synthetically generated data (70%). The synthetic dataset is derived from audio captioning datasets, where Large Language Models (LLMs) generate detailed and multiple-choice questions focusing on audio events, objects, acoustic scenes, signal properties, semantics, and listener emotions. To evaluate Mellow's reasoning ability, we benchmark it on a diverse set of tasks, assessing on both in-distribution and out-of-distribution data, including audio understanding, deductive reasoning, and comparative reasoning. Finally, we conduct extensive ablation studies to explore the impact of projection layer choices, synthetic data generation methods, and language model pretraining on reasoning performance. Our training dataset, findings, and baseline pave the way for developing small ALMs capable of reasoning.

While machine learning algorithms have been shown to excel at specific chemical tasks, they have struggled to capture the strategic thinking that characterizes expert chemical reasoning, limiting their widespread adoption. Here we demonstrate that large language models (LLMs) can serve as powerful chemical reasoning engines when integrated with traditional search algorithms, enabling a new approach to computer-aided chemistry that mirrors human expert thinking. Rather than using LLMs to directly manipulate chemical structures, we leverage their ability to evaluate chemical strategies and guide search algorithms toward chemically meaningful solutions. We demonstrate this paradigm through two fundamental challenges: strategy-aware retrosynthetic planning and mechanism elucidation. In retrosynthetic planning, our method allows chemists to specify desired synthetic strategies in natural language to find routes that satisfy these constraints in vast searches. In mechanism elucidation, LLMs guide the search for plausible reaction mechanisms by combining chemical principles with systematic exploration. Our approach shows strong performance across diverse chemical tasks, with larger models demonstrating increasingly sophisticated chemical reasoning. Our approach establishes a new paradigm for computer-aided chemistry that combines the strategic understanding of LLMs with the precision of traditional chemical tools, opening possibilities for more intuitive and powerful chemical reasoning systems.

Reinforcement learning with verifiable outcome rewards (RLVR) has effectively scaled up chain-of-thought (CoT) reasoning in large language models (LLMs). Yet, its efficacy in training vision-language model (VLM) agents for goal-directed action reasoning in visual environments is less established. This work investigates this problem through extensive experiments on complex card games, such as 24 points, and embodied tasks from ALFWorld. We find that when rewards are based solely on action outcomes, RL fails to incentivize CoT reasoning in VLMs, instead leading to a phenomenon we termed thought collapse, characterized by a rapid loss of diversity in the agent's thoughts, state-irrelevant and incomplete reasoning, and subsequent invalid actions, resulting in negative rewards. To counteract thought collapse, we highlight the necessity of process guidance and propose an automated corrector that evaluates and refines the agent's reasoning at each RL step. This simple and scalable GTR (Guided Thought Reinforcement) framework trains reasoning and action simultaneously without the need for dense, per-step human labeling. Our experiments demonstrate that GTR significantly enhances the performance and generalization of the LLaVA-7b model across various visual environments, achieving 3-5 times higher task success rates compared to SoTA models with notably smaller model sizes.

Due to the large number of parameters, the inference phase of Large Language Models (LLMs) is resource-intensive. Unlike traditional model compression, which needs retraining, recent dynamic computation methods show that not all components are required for inference, enabling a training-free pipeline. In this paper, we focus on the dynamic depth of LLM generation. A token-position aware layer skipping framework is proposed to save 1.5x times operations efficiently while maintaining performance. We first observed that tokens predicted later have lower perplexity and thus require less computation. Then, we propose a training-free algorithm called Position-Aware Depth Decay Decoding ($D^3$), which leverages a power-law decay function, $\left\lfloor L \times (\alpha^i) \right\rfloor$, to determine the number of layers to retain when generating token $T_i$. Remarkably, without any retraining, the $D^3$ achieves success across a wide range of generation tasks for the first time. Experiments on large language models (\ie the Llama) with $7 \sim 70$ billion parameters show that $D^3$ can achieve an average 1.5x speedup compared with the full-inference pipeline while maintaining comparable performance with nearly no performance drop ($<1\%$) on the GSM8K and BBH benchmarks.

Class-Incremental Learning (CIL) enables learning systems to continuously adapt to evolving data streams. With the advancement of pre-training, leveraging pre-trained vision-language models (e.g., CLIP) offers a promising starting point for CIL. However, CLIP makes decisions by matching visual embeddings to class names, overlooking the rich contextual information conveyed through language. For instance, the concept of ``cat'' can be decomposed into features like tail, fur, and face for recognition. Besides, since the model is continually updated, these detailed features are overwritten in CIL, requiring external knowledge for compensation. In this paper, we introduce ExterNal knowledGe INjEction (ENGINE) for CLIP-based CIL. To enhance knowledge transfer from outside the dataset, we propose a dual-branch injection tuning framework that encodes informative knowledge from both visual and textual modalities. The visual branch is enhanced with data augmentation to enrich the visual features, while the textual branch leverages GPT-4 to rewrite discriminative descriptors. In addition to this on-the-fly knowledge injection, we also implement post-tuning knowledge by re-ranking the prediction results during inference. With the injected knowledge, the model can better capture informative features for downstream tasks as data evolves. Extensive experiments demonstrate the state-of-the-art performance of ENGINE. Code is available at: https://github.com/RenaissCode/ENGINE

In recent years, lightweight large language models (LLMs) have garnered significant attention in the robotics field due to their low computational resource requirements and suitability for edge deployment. However, in task planning -- particularly for complex tasks that involve dynamic semantic logic reasoning -- lightweight LLMs have underperformed. To address this limitation, we propose a novel task planner, LightPlanner, which enhances the performance of lightweight LLMs in complex task planning by fully leveraging their reasoning capabilities. Unlike conventional planners that use fixed skill templates, LightPlanner controls robot actions via parameterized function calls, dynamically generating parameter values. This approach allows for fine-grained skill control and improves task planning success rates in complex scenarios. Furthermore, we introduce hierarchical deep reasoning. Before generating each action decision step, LightPlanner thoroughly considers three levels: action execution (feedback verification), semantic parsing (goal consistency verification), and parameter generation (parameter validity verification). This ensures the correctness of subsequent action controls. Additionally, we incorporate a memory module to store historical actions, thereby reducing context length and enhancing planning efficiency for long-term tasks. We train the LightPlanner-1.5B model on our LightPlan-40k dataset, which comprises 40,000 action controls across tasks with 2 to 13 action steps. Experiments demonstrate that our model achieves the highest task success rate despite having the smallest number of parameters. In tasks involving spatial semantic reasoning, the success rate exceeds that of ReAct by 14.9 percent. Moreover, we demonstrate LightPlanner's potential to operate on edge devices.