The preservation of aquatic biodiversity is critical in mitigating the effects of climate change. Aquatic scene understanding plays a pivotal role in aiding marine scientists in their decision-making processes. In this paper, we introduce AquaticCLIP, a novel contrastive language-image pre-training model tailored for aquatic scene understanding. AquaticCLIP presents a new unsupervised learning framework that aligns images and texts in aquatic environments, enabling tasks such as segmentation, classification, detection, and object counting. By leveraging our large-scale underwater image-text paired dataset without the need for ground-truth annotations, our model enriches existing vision-language models in the aquatic domain. For this purpose, we construct a 2 million underwater image-text paired dataset using heterogeneous resources, including YouTube, Netflix, NatGeo, etc. To fine-tune AquaticCLIP, we propose a prompt-guided vision encoder that progressively aggregates patch features via learnable prompts, while a vision-guided mechanism enhances the language encoder by incorporating visual context. The model is optimized through a contrastive pretraining loss to align visual and textual modalities. AquaticCLIP achieves notable performance improvements in zero-shot settings across multiple underwater computer vision tasks, outperforming existing methods in both robustness and interpretability. Our model sets a new benchmark for vision-language applications in underwater environments. The code and dataset for AquaticCLIP are publicly available on GitHub at xxx.

Federated multi-task learning (FMTL) aims to simultaneously learn multiple related tasks across clients without sharing sensitive raw data. However, in the decentralized setting, existing FMTL frameworks are limited in their ability to capture complex task relationships and handle feature and sample heterogeneity across clients. To address these challenges, we introduce a novel sheaf-theoretic-based approach for FMTL. By representing client relationships using cellular sheaves, our framework can flexibly model interactions between heterogeneous client models. We formulate the sheaf-based FMTL optimization problem using sheaf Laplacian regularization and propose the Sheaf-FMTL algorithm to solve it. We show that the proposed framework provides a unified view encompassing many existing federated learning (FL) and FMTL approaches. Furthermore, we prove that our proposed algorithm, Sheaf-FMTL, achieves a sublinear convergence rate in line with state-of-the-art decentralized FMTL algorithms. Extensive experiments demonstrate that Sheaf-FMTL exhibits communication savings by sending significantly fewer bits compared to decentralized FMTL baselines.

The availability of foundational models (FMs) pre-trained on large-scale data has advanced the state-of-the-art in many computer vision tasks. While FMs have demonstrated good zero-shot performance on many image classification tasks, there is often scope for performance improvement by adapting the FM to the downstream task. However, the data that is required for this adaptation typically exists in silos across multiple entities (data owners) and cannot be collated at a central location due to regulations and privacy concerns. At the same time, a learning service provider (LSP) who owns the FM cannot share the model with the data owners due to proprietary reasons. In some cases, the data owners may not even have the resources to store such large FMs. Hence, there is a need for algorithms to adapt the FM in a double-blind federated manner, i.e., the data owners do not know the FM or each other's data, and the LSP does not see the data for the downstream tasks. In this work, we propose a framework for double-blind federated adaptation of FMs using fully homomorphic encryption (FHE). The proposed framework first decomposes the FM into a sequence of FHE-friendly blocks through knowledge distillation. The resulting FHE-friendly model is adapted for the downstream task via low-rank parallel adapters that can be learned without backpropagation through the FM. Since the proposed framework requires the LSP to share intermediate representations with the data owners, we design a privacy-preserving permutation scheme to prevent the data owners from learning the FM through model extraction attacks. Finally, a secure aggregation protocol is employed for federated learning of the low-rank parallel adapters. Empirical results on four datasets demonstrate the practical feasibility of the proposed framework.

In this work, we explore how instance-level memorization in the teacher Neural Machine Translation (NMT) model gets inherited by the student model in sequence-level knowledge distillation (SeqKD). We find that despite not directly seeing the original training data, students memorize more than baseline models (models of the same size, trained on the original data) -- 3.4% for exact matches and 57% for extractive memorization -- and show increased hallucination rates. Further, under this SeqKD setting, we also characterize how students behave on specific training data subgroups, such as subgroups with low quality and specific counterfactual memorization (CM) scores, and find that students exhibit amplified denoising on low-quality subgroups. Finally, we propose a modification to SeqKD named Adaptive-SeqKD, which intervenes in SeqKD to reduce memorization and hallucinations. Overall, we recommend caution when applying SeqKD: students inherit both their teachers' superior performance and their fault modes, thereby requiring active monitoring.

Gradient Boosting Decision Tree (GBDT) is one of the most popular machine learning models in various applications. However, in the traditional settings, all data should be simultaneously accessed in the training procedure: it does not allow to add or delete any data instances after training. In this paper, we propose an efficient online learning framework for GBDT supporting both incremental and decremental learning. To the best of our knowledge, this is the first work that considers an in-place unified incremental and decremental learning on GBDT. To reduce the learning cost, we present a collection of optimizations for our framework, so that it can add or delete a small fraction of data on the fly. We theoretically show the relationship between the hyper-parameters of the proposed optimizations, which enables trading off accuracy and cost on incremental and decremental learning. The backdoor attack results show that our framework can successfully inject and remove backdoor in a well-trained model using incremental and decremental learning, and the empirical results on public datasets confirm the effectiveness and efficiency of our proposed online learning framework and optimizations.

Layer-wise preconditioning methods are a family of memory-efficient optimization algorithms that introduce preconditioners per axis of each layer's weight tensors. These methods have seen a recent resurgence, demonstrating impressive performance relative to entry-wise ("diagonal") preconditioning methods such as Adam(W) on a wide range of neural network optimization tasks. Complementary to their practical performance, we demonstrate that layer-wise preconditioning methods are provably necessary from a statistical perspective. To showcase this, we consider two prototypical models, linear representation learning and single-index learning, which are widely used to study how typical algorithms efficiently learn useful features to enable generalization. In these problems, we show SGD is a suboptimal feature learner when extending beyond ideal isotropic inputs $\mathbf{x} \sim \mathsf{N}(\mathbf{0}, \mathbf{I})$ and well-conditioned settings typically assumed in prior work. We demonstrate theoretically and numerically that this suboptimality is fundamental, and that layer-wise preconditioning emerges naturally as the solution. We further show that standard tools like Adam preconditioning and batch-norm only mildly mitigate these issues, supporting the unique benefits of layer-wise preconditioning.

Human-robot cooperative navigation is challenging in environments with incomplete information. We introduce CoNav-Maze, a simulated robotics environment where a robot navigates using local perception while a human operator provides guidance based on an inaccurate map. The robot can share its camera views to improve the operator's understanding of the environment. To enable efficient human-robot cooperation, we propose Information Gain Monte Carlo Tree Search (IG-MCTS), an online planning algorithm that balances autonomous movement and informative communication. Central to IG-MCTS is a neural human perception dynamics model that estimates how humans distill information from robot communications. We collect a dataset through a crowdsourced mapping task in CoNav-Maze and train this model using a fully convolutional architecture with data augmentation. User studies show that IG-MCTS outperforms teleoperation and instruction-following baselines, achieving comparable task performance with significantly less communication and lower human cognitive load, as evidenced by eye-tracking metrics.

Video meeting platforms display conversations linearly through transcripts or summaries. However, ideas during a meeting do not emerge linearly. We leverage LLMs to create dialogue maps in real time to help people visually structure and connect ideas. Balancing the need to reduce the cognitive load on users during the conversation while giving them sufficient control when using AI, we explore two system variants that encompass different levels of AI assistance. In Human-Map, AI generates summaries of conversations as nodes, and users create dialogue maps with the nodes. In AI-Map, AI produces dialogue maps where users can make edits. We ran a within-subject experiment with ten pairs of users, comparing the two MeetMap variants and a baseline. Users preferred MeetMap over traditional methods for taking notes, which aligned better with their mental models of conversations. Users liked the ease of use for AI-Map due to the low effort demands and appreciated the hands-on opportunity in Human-Map for sense-making.

Reinforcement learning (RL) offers a general approach for modeling and training AI agents, including human-AI interaction scenarios. In this paper, we propose SHARPIE (Shared Human-AI Reinforcement Learning Platform for Interactive Experiments) to address the need for a generic framework to support experiments with RL agents and humans. Its modular design consists of a versatile wrapper for RL environments and algorithm libraries, a participant-facing web interface, logging utilities, deployment on popular cloud and participant recruitment platforms. It empowers researchers to study a wide variety of research questions related to the interaction between humans and RL agents, including those related to interactive reward specification and learning, learning from human feedback, action delegation, preference elicitation, user-modeling, and human-AI teaming. The platform is based on a generic interface for human-RL interactions that aims to standardize the field of study on RL in human contexts.

Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs), particularly in tasks requiring complex multi-step reasoning. While dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards, such as training efficiency and credit assignment, this potential remains largely unrealized. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive, making them particularly vulnerable to reward hacking. To address these challenges, we propose PRIME (Process Reinforcement through IMplicit rEwards), which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards. PRIME combines well with various advantage functions and forgoes the dedicated reward model training phrase that existing approaches require, substantially reducing the development overhead. We demonstrate PRIME's effectiveness on competitional math and coding. Starting from Qwen2.5-Math-7B-Base, PRIME achieves a 15.1% average improvement across several key reasoning benchmarks over the SFT model. Notably, our resulting model, Eurus-2-7B-PRIME, surpasses Qwen2.5-Math-7B-Instruct on seven reasoning benchmarks with 10% of its training data.