This paper focuses on monolithic Multimodal Large Language Models (MLLMs), which integrate visual encoding and language decoding into a single model. Existing structures and pre-training strategies for monolithic MLLMs often suffer from unstable optimization and catastrophic forgetting. To address these challenges, our key idea is to embed a new visual parameter space into a pre-trained LLM, enabling stable learning of visual knowledge from noisy data via delta tuning. Based on this principle, we first introduce Mono-InternVL, an advanced monolithic MLLM that incorporates a set of visual experts through a multimodal mixture-of-experts architecture. In addition, we design an innovative Endogenous Visual Pre-training (EViP) for Mono-InternVL to maximize its visual capabilities via progressive learning. Mono-InternVL achieves competitive performance against existing MLLMs but also leads to relatively expensive data cost. Therefore, we further present Mono-InternVL-1.5, a cheaper and stronger monolithic MLLM equipped with an improved EViP (EViP++). EViP++ introduces additional visual attention experts to Mono-InternVL-1.5 and re-organizes the pre-training process in an efficient manner. During inference, it includes a fused CUDA kernel to speed up its MoE operations. With these designs, Mono-InternVL-1.5 significantly reduces training and inference costs, while still maintaining competitive performance with Mono-InternVL. To evaluate our approach, we conduct extensive experiments across 15 benchmarks. Results demonstrate that Mono-InternVL outperforms existing monolithic MLLMs on 12 out of 15 benchmarks, e.g., +114-point improvement over Emu3 on OCRBench. Compared to its modular counterpart, i.e., InternVL-1.5, Mono-InternVL-1.5 achieves similar multimodal performance while reducing first-token latency by up to 69%. Code and models are released at https://github.com/OpenGVLab/Mono-InternVL.

The growing ubiquity of artificial intelligence (AI), in particular large language models (LLMs), has profoundly altered the way in which learners gain knowledge and interact with learning material, with many claiming that AI positively influences their learning achievements. Despite this advancement, current AI tutoring systems face limitations associated with their reactive nature, often providing direct answers without encouraging deep reflection or incorporating structured pedagogical tools and strategies. This limitation is most apparent in the field of mathematics, in which AI tutoring systems remain underdeveloped. This research addresses the question: How can AI tutoring systems move beyond providing reactive assistance to enable structured, individualized, and tool-assisted learning experiences? We introduce a novel multi-agent AI tutoring platform that combines adaptive and personalized feedback, structured course generation, and textbook knowledge retrieval to enable modular, tool-assisted learning processes. This system allows students to learn new topics while identifying and targeting their weaknesses, revise for exams effectively, and practice on an unlimited number of personalized exercises. This article contributes to the field of artificial intelligence in education by introducing a novel platform that brings together pedagogical agents and AI-driven components, augmenting the field with modular and effective systems for teaching mathematics.

The increasing use of generative AI for resume screening is predicated on the assumption that it offers an unbiased alternative to biased human decision-making. However, this belief fails to address a critical question: are these AI systems fundamentally competent at the evaluative tasks they are meant to perform? This study investigates the question of competence through a two-part audit of eight major AI platforms. Experiment 1 confirmed complex, contextual racial and gender biases, with some models penalizing candidates merely for the presence of demographic signals. Experiment 2, which evaluated core competence, provided a critical insight: some models that appeared unbiased were, in fact, incapable of performing a substantive evaluation, relying instead on superficial keyword matching. This paper introduces the "Illusion of Neutrality" to describe this phenomenon, where an apparent lack of bias is merely a symptom of a model's inability to make meaningful judgments. This study recommends that organizations and regulators adopt a dual-validation framework, auditing AI hiring tools for both demographic bias and demonstrable competence to ensure they are both equitable and effective.

Team modeling remains a fundamental challenge at the intersection of Artificial Intelligence and the Social Sciences. Social Science research emphasizes the need to jointly model dynamics and relations, while practical applications demand unified models capable of inferring multiple team constructs simultaneously, providing interpretable insights and actionable recommendations to enhance team performance. However, existing works do not meet these practical demands. To bridge this gap, we present TRENN, a novel tempo-relational architecture that integrates: (i) an automatic temporal graph extractor, (ii) a tempo-relational encoder, (iii) a decoder for team construct prediction, and (iv) two complementary explainability modules. TRENN jointly captures relational and temporal team dynamics, providing a solid foundation for MT-TRENN, which extends TReNN by replacing the decoder with a multi-task head, enabling the model to learn shared Social Embeddings and simultaneously predict multiple team constructs, including Emergent Leadership, Leadership Style, and Teamwork components. Experimental results demonstrate that our approach significantly outperforms approaches that rely exclusively on temporal or relational information. Additionally, experimental evaluation has shown that the explainability modules integrated in MT-TRENN yield interpretable insights and actionable suggestions to support team improvement. These capabilities make our approach particularly well-suited for Human-Centered AI applications, such as intelligent decision-support systems in high-stakes collaborative environments.

A central question in computer science and statistics is whether efficient algorithms can achieve the information-theoretic limits of statistical problems. Many computational-statistical tradeoffs have been shown under average-case assumptions, but since statistical problems are average-case in nature, it has been a challenge to base them on standard worst-case assumptions. In PAC learning where such tradeoffs were first studied, the question is whether computational efficiency can come at the cost of using more samples than information-theoretically necessary. We base such tradeoffs on $\mathsf{NP}$-hardness and obtain: $\circ$ Sharp computational-statistical tradeoffs assuming $\mathsf{NP}$ requires exponential time: For every polynomial $p(n)$, there is an $n$-variate class $C$ with VC dimension $1$ such that the sample complexity of time-efficiently learning $C$ is $Θ(p(n))$. $\circ$ A characterization of $\mathsf{RP}$ vs. $\mathsf{NP}$ in terms of learning: $\mathsf{RP} = \mathsf{NP}$ iff every $\mathsf{NP}$-enumerable class is learnable with $O(\mathrm{VCdim}(C))$ samples in polynomial time. The forward implication has been known since (Pitt and Valiant, 1988); we prove the reverse implication. Notably, all our lower bounds hold against improper learners. These are the first $\mathsf{NP}$-hardness results for improperly learning a subclass of polynomial-size circuits, circumventing formal barriers of Applebaum, Barak, and Xiao (2008).

Tools extend the manipulation abilities of robots, much like they do for humans. Despite human expertise in tool manipulation, teaching robots these skills faces challenges. The complexity arises from the interplay of two simultaneous points of contact: one between the robot and the tool, and another between the tool and the environment. Tactile and proximity sensors play a crucial role in identifying these complex contacts. However, learning tool manipulation using these sensors remains challenging due to limited real-world data and the large sim-to-real gap. To address this, we propose a few-shot tool-use skill transfer framework using multimodal sensing. The framework involves pre-training the base policy to capture contact states common in tool-use skills in simulation and fine-tuning it with human demonstrations collected in the real-world target domain to bridge the domain gap. We validate that this framework enables teaching surface-following tasks using tools with diverse physical and geometric properties with a small number of demonstrations on the Franka Emika robot arm. Our analysis suggests that the robot acquires new tool-use skills by transferring the ability to recognise tool-environment contact relationships from pre-trained to fine-tuned policies. Additionally, combining proximity and tactile sensors enhances the identification of contact states and environmental geometry.

Data classification without access to labeled samples remains a challenging problem. It usually depends on an appropriately chosen distance between features, a topic addressed in metric learning. Recently, Huizing, Cantini and Peyré proposed to simultaneously learn optimal transport (OT) cost matrices between samples and features of the dataset. This leads to the task of finding positive eigenvectors of a certain nonlinear function that maps cost matrices to OT distances. Having this basic idea in mind, we consider both the algorithmic and the modeling part of unsupervised metric learning. First, we examine appropriate algorithms and their convergence. In particular, we propose to use the stochastic random function iteration algorithm and prove that it converges linearly for our setting, although our operators are not paracontractive as it was required for convergence so far. Second, we ask the natural question if the OT distance can be replaced by other distances. We show how Mahalanobis-like distances fit into our considerations. Further, we examine an approach via graph Laplacians. In contrast to the previous settings, we have just to deal with linear functions in the wanted matrices here, so that simple algorithms from linear algebra can be applied.

--Advancements in foundation models have made it possible to conduct applications in various downstream tasks. Especially, the new era has witnessed a remarkable capability to extend Large Language Models (LLMs) for tackling tasks of 3D scene understanding. Current methods rely heavily on 3D point clouds, but the 3D point cloud reconstruction of an indoor scene often results in information loss. Some textureless planes or repetitive patterns are prone to omission and manifest as voids within the reconstructed 3D point clouds. Besides, objects with complex structures tend to introduce distortion of details caused by misalignments between the captured images and the dense reconstructed point clouds. Based on these insights, we propose Argus, a novel 3D multimodal framework that leverages multi-view images for enhanced 3D scene understanding with LLMs. In general, Argus can be treated as a 3D Large Multimodal Foundation Model (3D-LMM) since it takes various modalities as input(text instructions, 2D multi-view images, and 3D point clouds) and expands the capability of LLMs to tackle 3D tasks. Argus involves fusing and integrating multi-view images and camera poses into view-as-scene features, which interact with the 3D features to create comprehensive and detailed 3D-aware scene embeddings. Our approach compensates for the information loss while reconstructing 3D point clouds and helps LLMs better understand the 3D world. Extensive experiments demonstrate that our method outperforms existing 3D-LMMs in various downstream tasks. NTRODUCTION Received 8 August 2024; revised 23 March 2025; accepted 12 June 2025. Yifan Xu is with School of Computer Science and Engineering, Bei-hang University, Beijing 100191, China, also with Beijing Digital Native Digital City Research Center, Beijing 100084, China (email: xudax-ian2001@gmail.com).

Scaling deep neural networks (NN) of reinforcement learning (RL) algorithms has been shown to enhance performance when feature extraction networks are used but the gained performance comes at the significant expense of increased computational and memory complexity. Neural network pruning methods have successfully addressed this challenge in supervised learning. However, their application to RL is underexplored. We propose an approach to integrate simultaneous training and pruning within advanced RL methods, in particular to RL algorithms enhanced by the Online Feature Extractor Network (OFENet). Our networks (XiNet) are trained to solve stochastic optimization problems over the RL networks' weights and the parameters of variational Bernoulli distributions for 0/1 Random Variables $ξ$ scaling each unit in the networks. The stochastic problem formulation induces regularization terms that promote convergence of the variational parameters to 0 when a unit contributes little to the performance. In this case, the corresponding structure is rendered permanently inactive and pruned from its network. We propose a cost-aware, sparsity-promoting regularization scheme, tailored to the DenseNet architecture of OFENets expressing the parameter complexity of involved networks in terms of the parameters of the RVs in these networks. Then, when matching this cost with the regularization terms, the many hyperparameters associated with them are automatically selected, effectively combining the RL objectives and network compression. We evaluate our method on continuous control benchmarks (MuJoCo) and the Soft Actor-Critic RL agent, demonstrating that OFENets can be pruned considerably with minimal loss in performance. Furthermore, our results confirm that pruning large networks during training produces more efficient and higher performing RL agents rather than training smaller networks from scratch.

We explore an ASR-free method for isolated word reading assessment in low-resource settings. Our few-shot approach compares input child speech to a small set of adult-provided reference templates. Inputs and templates are encoded using intermediate layers from large self-supervised learned (SSL) models. Using an Afrikaans child speech benchmark, we investigate design options such as discretising SSL features and barycentre averaging of the templates. Idealised experiments show reasonable performance for adults, but a substantial drop for child speech input, even with child templates. Despite the success of employing SSL representations in low-resource speech tasks, our work highlights the limitations of SSL representations for processing child data when used in a few-shot classification system.