With the development of Multimodal Large Language Models (MLLMs), numerous outstanding accomplishments have emerged within the open-source community. Due to the complexity of creating and training multimodal data pairs, it is still a computational and time-consuming process to build powerful MLLMs. In this work, we introduce Capybara-OMNI, an MLLM that trains in a lightweight and efficient manner and supports understanding text, image, video, and audio modalities. We present in detail the framework design, the data construction, and the training recipe, to develop an MLLM step-by-step to obtain competitive performance. We also provide exclusive benchmarks utilized in our experiments to show how to properly verify understanding capabilities across different modalities. Results show that by following our guidance, we can efficiently build an MLLM that achieves competitive performance among models of the same scale on various multimodal benchmarks. Additionally, to enhance the multimodal instruction following and conversational capabilities of the model, we further discuss how to train the chat version upon an MLLM understanding model, which is more in line with user habits for tasks like real-time interaction with humans. We publicly disclose the Capybara-OMNI model, along with its chat-based version. The disclosure includes both the model weights, a portion of the training data, and the inference codes, which are made available on GitHub.

Heterogeneous model fusion enhances the performance of LLMs by integrating the knowledge and capabilities of multiple structurally diverse models. However, existing approaches often rely solely on selecting the best output for each prompt from source models, which underutilizes their full potential due to limited source knowledge and results in sparse optimization signals. To address this limitation, we propose FuseRL, a novel two-stage framework comprising FuseSFT and FusePO to maximize the utilization of source LLMs. FuseSFT establishes a robust initialization by integrating the strengths of heterogeneous source models through weighted supervised fine-tuning (SFT) on diverse outputs for each prompt. FusePO optimizes weighted preferences based on the outputs of multiple source models to enable superior alignment performance. Extensive experiments demonstrate the effectiveness of our framework across various preference alignment methods, including RLOO, DPO, and SimPO. Using Llama-3.1-8B-Instruct as the target model, our approach achieves state-of-the-art performance among 8B LLMs on the AlpacaEval-2 and Arena-Hard benchmarks. Further analysis suggests that FuseSFT regularizes the training process to reduce overfitting, while FusePO introduces dense and diverse signals for preference optimization.

Large Language Models (LLMs) equipped with web search capabilities have demonstrated impressive potential for deep research tasks. However, current approaches predominantly rely on either manually engineered prompts (prompt engineering-based) with brittle performance or reinforcement learning within controlled Retrieval-Augmented Generation (RAG) environments (RAG-based) that fail to capture the complexities of real-world interaction. In this paper, we introduce DeepResearcher, the first comprehensive framework for end-to-end training of LLM-based deep research agents through scaling reinforcement learning (RL) in real-world environments with authentic web search interactions. Unlike RAG-based approaches that assume all necessary information exists within a fixed corpus, our method trains agents to navigate the noisy, unstructured, and dynamic nature of the open web. We implement a specialized multi-agent architecture where browsing agents extract relevant information from various webpage structures and overcoming significant technical challenges. Extensive experiments on open-domain research tasks demonstrate that DeepResearcher achieves substantial improvements of up to 28.9 points over prompt engineering-based baselines and up to 7.2 points over RAG-based RL agents. Our qualitative analysis reveals emergent cognitive behaviors from end-to-end RL training, including the ability to formulate plans, cross-validate information from multiple sources, engage in self-reflection to redirect research, and maintain honesty when unable to find definitive answers. Our results highlight that end-to-end training in real-world web environments is not merely an implementation detail but a fundamental requirement for developing robust research capabilities aligned with real-world applications. We release DeepResearcher at https://github.com/GAIR-NLP/DeepResearcher.

In a series of interviews, we're meeting some of the AAAI/SIGAI Doctoral Consortium participants to find out more about their research. In this latest interview, we hear from Amina Mević who is applying machine learning to semiconductor manufacturing. Find out more about her PhD research so far, what makes this field so interesting, and how she found the AAAI Doctoral Consortium experience. I am currently pursuing my PhD at the University of Sarajevo, Faculty of Electrical Engineering, Department of Computer Science and Informatics. My research is being carried out in collaboration with Infineon Technologies Austria as part of the Important Project of Common European Interest (IPCEI) in Microelectronics.

This article considers the problem of designing adaption and optimisation techniques for training quantum learning machines. To this end, the division algebra of quaternions is used to derive an effective model for representing computation and measurement operations on qubits. In turn, the derived model, serves as the foundation for formulating an adaptive learning problem on principal quantum learning units, thereby establishing quantum information processing units akin to that of neurons in classical approaches. Then, leveraging the modern HR-calculus, a comprehensive training framework for learning on quantum machines is developed. The quaternion-valued model accommodates mathematical tractability and establishment of performance criteria, such as convergence conditions.

Spectral algorithms leverage spectral regularization techniques to analyze and process data, providing a flexible framework for addressing supervised learning problems. To deepen our understanding of their performance in real-world scenarios where the distributions of training and test data may differ, we conduct a rigorous investigation into the convergence behavior of spectral algorithms under distribution shifts, specifically within the framework of reproducing kernel Hilbert spaces. Our study focuses on the case of covariate shift. In this scenario, the marginal distributions of the input data differ between the training and test datasets, while the conditional distribution of the output given the input remains unchanged. Under this setting, we analyze the generalization error of spectral algorithms and show that they achieve minimax optimality when the density ratios between the training and test distributions are uniformly bounded. However, we also identify a critical limitation: when the density ratios are unbounded, the spectral algorithms may become suboptimal. To address this limitation, we propose a weighted spectral algorithm that incorporates density ratio information into the learning process. Our theoretical analysis shows that this weighted approach achieves optimal capacity-independent convergence rates. Furthermore, by introducing a weight clipping technique, we demonstrate that the convergence rates of the weighted spectral algorithm can approach the optimal capacity-dependent convergence rates arbitrarily closely. This improvement resolves the suboptimality issue in unbounded density ratio scenarios and advances the state-of-the-art by refining existing theoretical results.

We investigate the challenge of establishing stochastic-like guarantees when sequentially learning from a stream of i.i.d. data that includes an unknown quantity of clean-label adversarial samples. We permit the learner to abstain from making predictions when uncertain. The regret of the learner is measured in terms of misclassification and abstention error, where we allow the learner to abstain for free on adversarial injected samples. This approach is based on the work of Goel, Hanneke, Moran, and Shetty from arXiv:2306.13119. We explore the methods they present and manage to correct inaccuracies in their argumentation. However, this approach is limited to the realizable setting, where labels are assigned according to some function $f^*$ from the hypothesis space $\mathcal{F}$. Based on similar arguments, we explore methods to make adaptations for the agnostic setting where labels are random. Introducing the notion of a clean-label adversary in the agnostic context, we are the first to give a theoretical analysis of a disagreement-based learner for thresholds, subject to a clean-label adversary with noise.

We develop two classes of variance-reduced fast operator splitting methods to approximate solutions of both finite-sum and stochastic generalized equations. Our approach integrates recent advances in accelerated fixed-point methods, co-hypomonotonicity, and variance reduction. First, we introduce a class of variance-reduced estimators and establish their variance-reduction bounds. This class covers both unbiased and biased instances and comprises common estimators as special cases, including SVRG, SAGA, SARAH, and Hybrid-SGD. Next, we design a novel accelerated variance-reduced forward-backward splitting (FBS) algorithm using these estimators to solve finite-sum and stochastic generalized equations. Our method achieves both $\mathcal{O}(1/k^2)$ and $o(1/k^2)$ convergence rates on the expected squared norm $\mathbb{E}[ \| G_{\lambda}x^k\|^2]$ of the FBS residual $G_{\lambda}$, where $k$ is the iteration counter. Additionally, we establish, for the first time, almost sure convergence rates and almost sure convergence of iterates to a solution in stochastic accelerated methods. Unlike existing stochastic fixed-point algorithms, our methods accommodate co-hypomonotone operators, which potentially include nonmonotone problems arising from recent applications. We further specify our method to derive an appropriate variant for each stochastic estimator -- SVRG, SAGA, SARAH, and Hybrid-SGD -- demonstrating that they achieve the best-known complexity for each without relying on enhancement techniques. Alternatively, we propose an accelerated variance-reduced backward-forward splitting (BFS) method, which attains similar convergence rates and oracle complexity as our FBS method. Finally, we validate our results through several numerical experiments and compare their performance.

Current sign language machine translation systems rely on recognizing hand movements, facial expressions and body postures, and natural language processing, to convert signs into text. Recent approaches use Transformer architectures to model long-range dependencies via positional encoding. However, they lack accuracy in recognizing fine-grained, short-range temporal dependencies between gestures captured at high frame rates. Moreover, their high computational complexity leads to inefficient training. To mitigate these issues, we propose an Adaptive Transformer (ADAT), which incorporates components for enhanced feature extraction and adaptive feature weighting through a gating mechanism to emphasize contextually relevant features while reducing training overhead and maintaining translation accuracy. To evaluate ADAT, we introduce MedASL, the first public medical American Sign Language dataset. In sign-to-gloss-to-text experiments, ADAT outperforms the encoder-decoder transformer, improving BLEU-4 accuracy by 0.1% while reducing training time by 14.33% on PHOENIX14T and 3.24% on MedASL. In sign-to-text experiments, it improves accuracy by 8.7% and reduces training time by 2.8% on PHOENIX14T and achieves 4.7% higher accuracy and 7.17% faster training on MedASL. Compared to encoder-only and decoder-only baselines in sign-to-text, ADAT is at least 6.8% more accurate despite being up to 12.1% slower due to its dual-stream structure.

This invited paper introduces the concept of "proof-carrying neuro-symbolic code" and explains its meaning and value, from both the "neural" and the "symbolic" perspectives. The talk outlines the first successes and challenges that this new area of research faces. Keywords: Neural Networks Cyber-Physical System Verification Programming Languages Neuro-Symbolic Programs. 1 Neuro-Symbolic Proofs and Programs Proof Carrying Code is a long tradition within programming language research, broadly referring to methods that interleave verification with executable code, thus avoiding the inevitable discrepancies that arise when the code and the proofs are handled in different languages. Although the term was coined by Necula [50] almost three decades ago, with time, it grew to encompass any languages that are powerful enough to handle both the coding and the proving. Examples are dependently-typed (Agda, Idris, Coq/Rocq) and refinement-typed (F*, Liquid Haskell) languages.