FocalPO: Enhancing Preference Optimizing by Focusing on Correct Preference Rankings

Efficient preference optimization algorithms such as Direct Preference Optimization (DPO) have become a popular approach in aligning large language models (LLMs) with human preferences. These algorithms implicitly treat the LLM as a reward model, and focus on training it to correct misranked preference pairs. However, recent work~\citep{chen2024preference} empirically finds that DPO training \textit{rarely improves these misranked preference pairs}, despite its gradient emphasizing on these cases. We introduce FocalPO, a DPO variant that instead \textit{down-weighs} misranked preference pairs and prioritizes enhancing the model's understanding of pairs that it can already rank correctly. Inspired by Focal Loss used in vision tasks, FocalPO achieves this by adding a modulating factor to dynamically scale DPO loss. Our experiment demonstrates that FocalPO surpasses DPO and its variants on popular benchmarks like Alpaca Eval 2.0 using Mistral-Base-7B and Llama-3-Instruct-8B. Additionally, we empirically reveals how FocalPO affects training on correct and incorrect sample groups, further underscoring its effectiveness.

Memory-Reduced Meta-Learning with Guaranteed Convergence

The optimization-based meta-learning approach is gaining increased traction because of its unique ability to quickly adapt to a new task using only small amounts of data. However, existing optimization-based meta-learning approaches, such as MAML, ANIL and their variants, generally employ backpropagation for upper-level gradient estimation, which requires using historical lower-level parameters/gradients and thus increases computational and memory overhead in each iteration. In this paper, we propose a meta-learning algorithm that can avoid using historical parameters/gradients and significantly reduce memory costs in each iteration compared to existing optimization-based meta-learning approaches. In addition to memory reduction, we prove that our proposed algorithm converges sublinearly with the iteration number of upper-level optimization, and the convergence error decays sublinearly with the batch size of sampled tasks. In the specific case in terms of deterministic meta-learning, we also prove that our proposed algorithm converges to an exact solution. Moreover, we quantify that the computational complexity of the algorithm is on the order of $\mathcal{O}(\epsilon^{-1})$, which matches existing convergence results on meta-learning even without using any historical parameters/gradients. Experimental results on meta-learning benchmarks confirm the efficacy of our proposed algorithm.

HumanEval-V: Evaluating Visual Understanding and Reasoning Abilities of Large Multimodal Models Through Coding Tasks

Coding tasks have been valuable for evaluating Large Language Models (LLMs), as they demand the comprehension of high-level instructions, complex reasoning, and the implementation of functional programs -- core capabilities for advancing Artificial General Intelligence. Despite the progress in Large Multimodal Models (LMMs), which extend LLMs with visual perception and understanding capabilities, there remains a notable lack of coding benchmarks that rigorously assess these models, particularly in tasks that emphasize visual reasoning. To address this gap, we introduce HumanEval-V, a novel and lightweight benchmark specifically designed to evaluate LMMs' visual understanding and reasoning capabilities through code generation. HumanEval-V includes 108 carefully crafted, entry-level Python coding tasks derived from platforms like CodeForces and Stack Overflow. Each task is adapted by modifying the context and algorithmic patterns of the original problems, with visual elements redrawn to ensure distinction from the source, preventing potential data leakage. LMMs are required to complete the code solution based on the provided visual context and a predefined Python function signature outlining the task requirements. Every task is equipped with meticulously handcrafted test cases to ensure a thorough and reliable evaluation of model-generated solutions. We evaluate 19 state-of-the-art LMMs using HumanEval-V, uncovering significant challenges. Proprietary models like GPT-4o achieve only 13% pass@1 and 36.4% pass@10, while open-weight models with 70B parameters score below 4% pass@1. Ablation studies further reveal the limitations of current LMMs in vision reasoning and coding capabilities. These results underscore key areas for future research to enhance LMMs' capabilities. We have open-sourced our code and benchmark at https://github.com/HumanEval-V/HumanEval-V-Benchmark.

Teaching AI Agents to Search with Reflective-MCTS and Exploratory Learning

Autonomous agents have demonstrated significant potential in automating complex multistep decision-making tasks. However, even state-of-the-art vision-language models (VLMs), such as GPT-4o, still fall short of human-level performance, particularly in intricate web environments and long-horizon planning tasks. To address these limitations, we present Reflective Monte Carlo Tree Search (R-MCTS) and Exploratory Learning to build o1-like models for agentic applications. We first introduce R-MCTS, a novel test-time algorithm designed to enhance the ability of AI agents to explore decision space on the fly. R-MCTS extends traditional MCTS by 1) incorporating contrastive reflection, allowing agents to learn from past interactions and dynamically improve their search efficiency; and 2) using multi-agent debate to provide reliable state evaluation. Next, we introduce Exploratory Learning, a novel learning strategy to teach agents to search at inference time without relying on any external search algorithms. On the challenging VisualWebArena benchmark, our GPT-4o-based R-MCTS agent achieves a 6% to 30% relative improvement across various tasks compared to the previous state-of-the-art. Additionally, we show that the experience gained from test-time search can be effectively transferred back to GPT-4o via fine-tuning. After Exploratory Learning, GPT-4o 1) demonstrates the ability to explore the environment, evaluate a state, and backtrack to viable ones when it detects that the current state cannot lead to success, and 2) matches 87% of R-MCTS's performance while using significantly less compute. Notably, our work demonstrates the compute scaling properties in both training - data collection with R-MCTS - and testing time. These results suggest a promising research direction to enhance VLMs' reasoning and planning capabilities for agentic applications via test-time search and self-learning.

Improving Autonomous AI Agents with Reflective Tree Search and Self-Learning

Autonomous agents have demonstrated significant potential in automating complex multistep decision-making tasks. However, even state-of-the-art vision-language models (VLMs), such as GPT-4o, still fall short of human-level performance, particularly in intricate web environments and long-horizon planning tasks. To address these limitations, we introduce Reflective Monte Carlo Tree Search (R-MCTS), a novel test-time algorithm designed to enhance the ability of AI agents, e.g., powered by GPT-4o, to explore decision space on the fly. R-MCTS extends traditional MCTS by 1) incorporating contrastive reflection, allowing agents to learn from past interactions and dynamically improve their search efficiency; and 2) using multi-agent debate to provide reliable state evaluation. Moreover, we improve the agent's performance by fine-tuning GPT-4o through self-learning, using R-MCTS generated tree traversals without any human-provided labels. On the challenging VisualWebArena benchmark, our GPT-4o-based R-MCTS agent achieves a 6% to 30% relative improvement across various tasks compared to the previous state-of-the-art. Additionally, we show that the knowledge gained from test-time search can be effectively transferred back to GPT-4o via fine-tuning. The fine-tuned GPT-4o matches 97% of R-MCTS's performance while reducing compute usage by a factor of four at test time. Furthermore, qualitative results reveal that the fine-tuned GPT-4o model demonstrates the ability to explore the environment, evaluate a state, and backtrack to viable ones when it detects that the current state cannot lead to success. Moreover, our work demonstrates the compute scaling properties in both training - data collection with R-MCTS - and testing time. These results suggest a promising research direction to enhance VLMs' reasoning and planning capabilities for agentic applications via test-time search and self-learning.

LPUWF-LDM: Enhanced Latent Diffusion Model for Precise Late-phase UWF-FA Generation on Limited Dataset

Ultra-Wide-Field Fluorescein Angiography (UWF-FA) enables precise identification of ocular diseases using sodium fluorescein, which can be potentially harmful. Existing research has developed methods to generate UWF-FA from Ultra-Wide-Field Scanning Laser Ophthalmoscopy (UWF-SLO) to reduce the adverse reactions associated with injections. However, these methods have been less effective in producing high-quality late-phase UWF-FA, particularly in lesion areas and fine details. Two primary challenges hinder the generation of high-quality late-phase UWF-FA: the scarcity of paired UWF-SLO and early/late-phase UWF-FA datasets, and the need for realistic generation at lesion sites and potential blood leakage regions. This study introduces an improved latent diffusion model framework to generate high-quality late-phase UWF-FA from limited paired UWF images. To address the challenges as mentioned earlier, our approach employs a module utilizing Cross-temporal Regional Difference Loss, which encourages the model to focus on the differences between early and late phases. Additionally, we introduce a low-frequency enhanced noise strategy in the diffusion forward process to improve the realism of medical images. To further enhance the mapping capability of the variational autoencoder module, especially with limited datasets, we implement a Gated Convolutional Encoder to extract additional information from conditional images. Our Latent Diffusion Model for Ultra-Wide-Field Late-Phase Fluorescein Angiography (LPUWF-LDM) effectively reconstructs fine details in late-phase UWF-FA and achieves state-of-the-art results compared to other existing methods when working with limited datasets. Our source code is available at: https://github.com/Tinysqua/****.

TC-KANRecon: High-Quality and Accelerated MRI Reconstruction via Adaptive KAN Mechanisms and Intelligent Feature Scaling

Magnetic Resonance Imaging (MRI) has become essential in clinical diagnosis due to its high resolution and multiple contrast mechanisms. However, the relatively long acquisition time limits its broader application. To address this issue, this study presents an innovative conditional guided diffusion model, named as TC-KANRecon, which incorporates the Multi-Free U-KAN (MF-UKAN) module and a dynamic clipping strategy. TC-KANRecon model aims to accelerate the MRI reconstruction process through deep learning methods while maintaining the quality of the reconstructed images. The MF-UKAN module can effectively balance the tradeoff between image denoising and structure preservation. Specifically, it presents the multi-head attention mechanisms and scalar modulation factors, which significantly enhances the model's robustness and structure preservation capabilities in complex noise environments. Moreover, the dynamic clipping strategy in TC-KANRecon adjusts the cropping interval according to the sampling steps, thereby mitigating image detail loss typically caused by traditional cropping methods and enriching the visual features of the images. Furthermore, the MC-Model module incorporates full-sampling k-space information, realizing efficient fusion of conditional information, enhancing the model's ability to process complex data, and improving the realism and detail richness of reconstructed images. Experimental results demonstrate that the proposed method outperforms other MRI reconstruction methods in both qualitative and quantitative evaluations. Notably, TC-KANRecon method exhibits excellent reconstruction results when processing high-noise, low-sampling-rate MRI data. Our source code is available at https://github.com/lcbkmm/TC-KANRecon.

LIONs: An Empirically Optimized Approach to Align Language Models

Alignment is a crucial step to enhance the instruction-following and conversational abilities of language models. Despite many recent work proposing new algorithms, datasets, and training pipelines, there is a lack of comprehensive studies measuring the impact of various design choices throughout the whole training process. We first conduct a rigorous analysis over a three-stage training pipeline consisting of supervised fine-tuning, offline preference learning, and online preference learning. We have found that using techniques like sequence packing, loss masking in SFT, increasing the preference dataset size in DPO, and online DPO training can significantly improve the performance of language models. We then train from Gemma-2b-base and LLama-3-8b-base, and find that our best models exceed the performance of the official instruct models tuned with closed-source data and algorithms. Our code and models can be found at https://github.com/Columbia-NLP-Lab/LionAlignment.

Diffusion Synthesizer for Efficient Multilingual Speech to Speech Translation

We introduce DiffuseST, a low-latency, direct speech-to-speech translation system capable of preserving the input speaker's voice zero-shot while translating from multiple source languages into English. We experiment with the synthesizer component of the architecture, comparing a Tacotron-based synthesizer to a novel diffusion-based synthesizer. We find the diffusion-based synthesizer to improve MOS and PESQ audio quality metrics by 23\% each and speaker similarity by 5\% while maintaining comparable BLEU scores. Despite having more than double the parameter count, the diffusion synthesizer has lower latency, allowing the entire model to run more than 5$\times$ faster than real-time.

UVEB: A Large-scale Benchmark and Baseline Towards Real-World Underwater Video Enhancement

Learning-based underwater image enhancement (UIE) methods have made great progress. However, the lack of large-scale and high-quality paired training samples has become the main bottleneck hindering the development of UIE. The inter-frame information in underwater videos can accelerate or optimize the UIE process. Thus, we constructed the first large-scale high-resolution underwater video enhancement benchmark (UVEB) to promote the development of underwater vision.It contains 1,308 pairs of video sequences and more than 453,000 high-resolution with 38\% Ultra-High-Definition (UHD) 4K frame pairs. UVEB comes from multiple countries, containing various scenes and video degradation types to adapt to diverse and complex underwater environments. We also propose the first supervised underwater video enhancement method, UVE-Net. UVE-Net converts the current frame information into convolutional kernels and passes them to adjacent frames for efficient inter-frame information exchange. By fully utilizing the redundant degraded information of underwater videos, UVE-Net completes video enhancement better. Experiments show the effective network design and good performance of UVE-Net.