Embodied Question Answering (EQA) has primarily focused on indoor environments, leaving the complexities of urban settings - spanning environment, action, and perception - largely unexplored. To bridge this gap, we introduce CityEQA, a new task where an embodied agent answers open-vocabulary questions through active exploration in dynamic city spaces. To support this task, we present CityEQA-EC, the first benchmark dataset featuring 1,412 human-annotated tasks across six categories, grounded in a realistic 3D urban simulator. Moreover, we propose Planner-Manager-Actor (PMA), a novel agent tailored for CityEQA. PMA enables long-horizon planning and hierarchical task execution: the Planner breaks down the question answering into sub-tasks, the Manager maintains an object-centric cognitive map for spatial reasoning during the process control, and the specialized Actors handle navigation, exploration, and collection sub-tasks. Experiments demonstrate that PMA achieves 60.7% of human-level answering accuracy, significantly outperforming frontier-based baselines. While promising, the performance gap compared to humans highlights the need for enhanced visual reasoning in CityEQA. This work paves the way for future advancements in urban spatial intelligence. Dataset and code are available at https://github.com/BiluYong/CityEQA.git.

Large language models (LLMs) have achieved significant performance gains via scaling up model sizes and/or data. However, recent evidence suggests diminishing returns from such approaches, motivating scaling the computation spent at inference time. Existing inference-time scaling methods, usually with reward models, cast the task as a search problem, which tends to be vulnerable to reward hacking as a consequence of approximation errors in reward models. In this paper, we instead cast inference-time scaling as a probabilistic inference task and leverage sampling-based techniques to explore the typical set of the state distribution of a state-space model with an approximate likelihood, rather than optimize for its mode directly. We propose a novel inference-time scaling approach by adapting particle-based Monte Carlo methods to this task. Our empirical evaluation demonstrates that our methods have a 4-16x better scaling rate over our deterministic search counterparts on various challenging mathematical reasoning tasks. Using our approach, we show that Qwen2.5-Math-1.5B-Instruct can surpass GPT-4o accuracy in only 4 rollouts, while Qwen2.5-Math-7B-Instruct scales to o1 level accuracy in only 32 rollouts. Our work not only presents an effective method to inference-time scaling, but also connects the rich literature in probabilistic inference with inference-time scaling of LLMs to develop more robust algorithms in future work. Code, videos, and further information available at https://probabilistic-inference-scaling.github.io.

The rise of large language models (LLMs) has created a significant disparity: industrial research labs with their computational resources, expert teams, and advanced infrastructures, can effectively fine-tune LLMs, while individual developers and small organizations face barriers due to limited resources. In this paper, we aim to bridge this gap by presenting a comprehensive study on supervised fine-tuning of LLMs using instruction-tuning datasets spanning diverse knowledge domains and skills. We focus on small-sized LLMs (3B to 7B parameters) for their cost-efficiency and accessibility. We explore various training configurations and strategies across four open-source pre-trained models. We provide detailed documentation of these configurations, revealing findings that challenge several common training practices, including hyperparameter recommendations from TULU and phased training recommended by Orca. Key insights from our work include: (i) larger batch sizes paired with lower learning rates lead to improved model performance on benchmarks such as MMLU, MTBench, and Open LLM Leaderboard; (ii) early-stage training dynamics, such as lower gradient norms and higher loss values, are strong indicators of better final model performance, enabling early termination of sub-optimal runs and significant computational savings; (iii) through a thorough exploration of hyperparameters like warmup steps and learning rate schedules, we provide guidance for practitioners and find that certain simplifications do not compromise performance; and (iv) we observed no significant difference in performance between phased and stacked training strategies, but stacked training is simpler and more sample efficient. With these findings holding robustly across datasets and models, we hope this study serves as a guide for practitioners fine-tuning small LLMs and promotes a more inclusive environment for LLM research.

Exploratory data analysis (EDA), coupled with SQL, is essential for data analysts involved in data exploration and analysis. However, data analysts often encounter two primary challenges: (1) the need to craft SQL queries skillfully, and (2) the requirement to generate suitable visualization types that enhance the interpretation of query results. Due to its significance, substantial research efforts have been made to explore different approaches to address these challenges, including leveraging large language models (LLMs). However, existing methods fail to meet real-world data exploration requirements primarily due to (1) complex database schema; (2) unclear user intent; (3) limited cross-domain generalization capability; and (4) insufficient end-to-end text-to-visualization capability. This paper presents TiInsight, an automated SQL-based cross-domain exploratory data analysis system. First, we propose hierarchical data context (i.e., HDC), which leverages LLMs to summarize the contexts related to the database schema, which is crucial for open-world EDA systems to generalize across data domains. Second, the EDA system is divided into four components (i.e., stages): HDC generation, question clarification and decomposition, text-to-SQL generation (i.e., TiSQL), and data visualization (i.e., TiChart). Finally, we implemented an end-to-end EDA system with a user-friendly GUI interface in the production environment at PingCAP. We have also open-sourced all APIs of TiInsight to facilitate research within the EDA community. Through extensive evaluations by a real-world user study, we demonstrate that TiInsight offers remarkable performance compared to human experts. Specifically, TiSQL achieves an execution accuracy of 86.3% on the Spider dataset using GPT-4. It also demonstrates state-of-the-art performance on the Bird dataset.

Object goal navigation (ObjectNav) is a fundamental task of embodied AI that requires the agent to find a target object in unseen environments. This task is particularly challenging as it demands both perceptual and cognitive processes for effective perception and decision-making. While perception has gained significant progress powered by the rapidly developed visual foundation models, the progress on the cognitive side remains limited to either implicitly learning from massive navigation demonstrations or explicitly leveraging pre-defined heuristic rules. Inspired by neuroscientific evidence that humans consistently update their cognitive states while searching for objects in unseen environments, we present CogNav, which attempts to model this cognitive process with the help of large language models. Specifically, we model the cognitive process with a finite state machine composed of cognitive states ranging from exploration to identification. The transitions between the states are determined by a large language model based on an online built heterogeneous cognitive map containing spatial and semantic information of the scene being explored. Extensive experiments on both synthetic and real-world environments demonstrate that our cognitive modeling significantly improves ObjectNav efficiency, with human-like navigation behaviors. In an open-vocabulary and zero-shot setting, our method advances the SOTA of the HM3D benchmark from 69.3% to 87.2%. The code and data will be released.

Preference tuning of large language models (LLMs) relies on high-quality human preference data, which is often expensive and time-consuming to gather. While existing methods can use trained reward models or proprietary model as judges for preference annotation, they have notable drawbacks: training reward models remain dependent on initial human data, and using proprietary model imposes license restrictions that inhibits commercial usage. In this paper, we introduce customized density ratio (CDR), a training-free and highly effective method that leverages off-the-shelf LLMs for preference data annotation. Our approach uses the log-density ratio between a better-aligned LLM and a less aligned LLM as a reward signal. We explores 221 different LLMs pairs and empirically demonstrate that increasing the performance gap between paired LLMs correlates with better reward generalization. Furthermore, we show that tailoring the density ratio reward function with specific criteria and preference exemplars enhances performance across domains and within target areas. In our experiment using density ratio from a pair of Mistral-7B models, CDR achieves a RewardBench score of 82.6, outperforming the best trained reward functions from same model class and demonstrating competitive performance against SoTA models in Safety (91.0) and Reasoning (88.0) domains. We use CDR to annotate an on-policy preference dataset with which we preference tune Llama-3-8B-Instruct with SimPO. Using reward signals from two relatively weak models, our approach pushes Llama-3-8B to achieve a 37.4% (+15.1%) Preference tuning has advanced the capabilities of large language models (LLMs), but this progress relies on high-quality human preference data which is both costly and time-consuming to gather. Cutting-edge models (e.g., ChatGPT, GPT-4, Claude-3) are aligned with curated, quality-controlled human preference data, typically provided by specialized companies. AI-feedback solutions are emerging as an alternative--either through a trained reward model (Dong et al., 2024) or proprietary LLM-as-a-judge (Cui et al., 2023). However, training reward models still rely on costly initial human preference data, and proprietary LLM-as-a-judge approaches introduce licensing restrictions that generally prevent commercial use.

Training generative models with differential privacy (DP) typically involves injecting noise into gradient updates or adapting the discriminator's training procedure. As a result, such approaches often struggle with hyper-parameter tuning and convergence. We consider the slicing privacy mechanism that injects noise into random low-dimensional projections of the private data, and provide strong privacy guarantees for it. These noisy projections are used for training generative models. To enable optimizing generative models using this DP approach, we introduce the smoothed-sliced $f$-divergence and show it enjoys statistical consistency. Moreover, we present a kernel-based estimator for this divergence, circumventing the need for adversarial training. Extensive numerical experiments demonstrate that our approach can generate synthetic data of higher quality compared with baselines. Beyond performance improvement, our method, by sidestepping the need for noisy gradients, offers data scientists the flexibility to adjust generator architecture and hyper-parameters, run the optimization over any number of epochs, and even restart the optimization process -- all without incurring additional privacy costs.

Fourier features based positional encoding (PE) is commonly used in machine learning tasks that involve learning high-frequency features from low-dimensional inputs, such as 3D view synthesis and time series regression with neural tangent kernels. Despite their effectiveness, existing PEs require manual, empirical adjustment of crucial hyperparameters, specifically the Fourier features, tailored to each unique task. Further, PEs face challenges in efficiently learning high-frequency functions, particularly in tasks with limited data. In this paper, we introduce sinusoidal PE (SPE), designed to efficiently learn adaptive frequency features closely aligned with the true underlying function. Our experiments demonstrate that SPE, without hyperparameter tuning, consistently achieves enhanced fidelity and faster training across various tasks, including 3D view synthesis, Text-to-Speech generation, and 1D regression. SPE is implemented as a direct replacement for existing PEs. Its plug-and-play nature lets numerous tasks easily adopt and benefit from SPE.

Reaching-and-grasping is a fundamental skill for robotic manipulation, but existing methods usually train models on a specific gripper and cannot be reused on another gripper. In this paper, we propose a novel method that can learn a unified policy model that can be easily transferred to different dexterous grippers. Our method consists of two stages: a gripper-agnostic policy model that predicts the displacements of pre-defined key points on the gripper, and a gripper-specific adaptation model that translates these displacements into adjustments for controlling the grippers' joints. The gripper state and interactions with objects are captured at the finger level using robust geometric representations, integrated with a transformer-based network to address variations in gripper morphology and geometry. In the experiments, we evaluate our method on several dexterous grippers and diverse objects, and the result shows that our method significantly outperforms the baseline methods. Pioneering the transfer of grasp policies across dexterous grippers, our method effectively demonstrates its potential for learning generalizable and transferable manipulation skills for various robotic hands.

To effectively implement quantum algorithms on noisy intermediate-scale quantum (NISQ) processors is a central task in modern quantum technology. NISQ processors feature tens to a few hundreds of noisy qubits with limited coherence times and gate operations with errors, so NISQ algorithms naturally require employing circuits of short lengths via quantum compilation. Here, we develop a reinforcement learning (RL)-based quantum compiler for a superconducting processor and demonstrate its capability of discovering novel and hardware-amenable circuits with short lengths. We show that for the three-qubit quantum Fourier transformation, a compiled circuit using only seven CZ gates with unity circuit fidelity can be achieved. The compiler is also able to find optimal circuits under device topological constraints, with lengths considerably shorter than those by the conventional method. Our study exemplifies the codesign of the software with hardware for efficient quantum compilation, offering valuable insights for the advancement of RL-based compilers.