Federated learning (FL) enhances privacy by keeping user data on local devices. However, emerging attacks have demonstrated that the updates shared by users during training can reveal significant information about their data. This has greatly thwart the adoption of FL methods for training robust AI models in sensitive applications. Differential Privacy (DP) is considered the gold standard for safeguarding user data. However, DP guarantees are highly conservative, providing worst-case privacy guarantees. This can result in overestimating privacy needs, which may compromise the model's accuracy. Additionally, interpretations of these privacy guarantees have proven to be challenging in different contexts. This is further exacerbated when other factors, such as the number of training iterations, data distribution, and specific application requirements, can add further complexity to this problem. In this work, we proposed a framework that integrates a human entity as a privacy practitioner to determine an optimal trade-off between the model's privacy and utility. Our framework is the first to address the variable memory requirement of existing DP methods in FL settings, where resource-limited devices (e.g., cell phones) can participate. To support such settings, we adopt a recent DP method with fixed memory usage to ensure scalable private FL. We evaluated our proposed framework by fine-tuning a BERT-based LLM model using the GLUE dataset (a common approach in literature), leveraging the new accountant, and employing diverse data partitioning strategies to mimic real-world conditions. As a result, we achieved stable memory usage, with an average accuracy reduction of 1.33% for $\epsilon = 10$ and 1.9% for $\epsilon = 6$, when compared to the state-of-the-art DP accountant which does not support fixed memory usage.

Recent advancements in Large Vision Language Models (LVLMs) have enabled the development of LVLM-based Graphical User Interface (GUI) agents under various paradigms. Training-based approaches, such as CogAgent and SeeClick, struggle with cross-dataset and cross-platform generalization due to their reliance on dataset-specific training. Generalist LVLMs, such as GPT-4V, employ Set-of-Marks (SoM) for action grounding, but obtaining SoM labels requires metadata like HTML source, which is not consistently available across platforms. Moreover, existing methods often specialize in singular GUI tasks rather than achieving comprehensive GUI understanding. To address these limitations, we introduce TRISHUL, a novel, training-free agentic framework that enhances generalist LVLMs for holistic GUI comprehension. Unlike prior works that focus on either action grounding (mapping instructions to GUI elements) or GUI referring (describing GUI elements given a location), TRISHUL seamlessly integrates both. At its core, TRISHUL employs Hierarchical Screen Parsing (HSP) and the Spatially Enhanced Element Description (SEED) module, which work synergistically to provide multi-granular, spatially, and semantically enriched representations of GUI elements. Our results demonstrate TRISHUL's superior performance in action grounding across the ScreenSpot, VisualWebBench, AITW, and Mind2Web datasets. Additionally, for GUI referring, TRISHUL surpasses the ToL agent on the ScreenPR benchmark, setting a new standard for robust and adaptable GUI comprehension.

As large language models (LLMs) advance, their ability to perform in-context learning and few-shot language generation has improved significantly. This has spurred using LLMs to produce high-quality synthetic data to enhance the performance of smaller models like online retrievers or weak LLMs. However, LLM-generated synthetic data often differs from the real data in key language attributes (e.g., styles, tones, content proportions, etc.). As a result, mixing these synthetic data directly with real data may distort the original data distribution, potentially hindering performance improvements. To solve this, we introduce SynAlign: a synthetic data generation and filtering framework based on key attribute distribution matching. Before generation, SynAlign employs an uncertainty tracker surrogated by the Gaussian Process model to iteratively select data clusters distinct from selected ones as demonstrations for new data synthesis, facilitating the efficient exploration diversity of the real data. Then, a latent attribute reasoning method is employed: the LLM summarizes linguistic attributes of demonstrations and then synthesizes new data based on them. This approach facilitates synthesizing diverse data with linguistic attributes that appear in real data.After generation, the Maximum Mean Discrepancy is used as the objective function to learn the sampling weight of each synthetic data, ensuring distribution matching with the real data. Our experiments on multiple text prediction tasks show significant performance improvements. We also conducted an online A/B test on an online retriever to demonstrate SynAlign's effectiveness.

As language models master existing reasoning benchmarks, we need new challenges to evaluate their cognitive frontiers. Puzzle-solving events are rich repositories of challenging multimodal problems that test a wide range of advanced reasoning and knowledge capabilities, making them a unique testbed for evaluating frontier language models. We introduce EnigmaEval, a dataset of problems and solutions derived from puzzle competitions and events that probes models' ability to perform implicit knowledge synthesis and multi-step deductive reasoning. Unlike existing reasoning and knowledge benchmarks, puzzle solving challenges models to discover hidden connections between seemingly unrelated pieces of information to uncover solution paths. The benchmark comprises 1184 puzzles of varying complexity -- each typically requiring teams of skilled solvers hours to days to complete -- with unambiguous, verifiable solutions that enable efficient evaluation. State-of-the-art language models achieve extremely low accuracy on these puzzles, even lower than other difficult benchmarks such as Humanity's Last Exam, unveiling models' shortcomings when challenged with problems requiring unstructured and lateral reasoning.

Large language models (LLMs) have demonstrated significant utilities in real-world applications, exhibiting impressive capabilities in natural language processing and understanding. Benchmark evaluations are crucial for assessing the capabilities of LLMs as they can provide a comprehensive assessment of their strengths and weaknesses. However, current evaluation methods often overlook the inherent randomness of LLMs by employing deterministic generation strategies or relying on a single random sample, resulting in unaccounted sampling variance and unreliable benchmark score estimates. In this paper, we propose a hierarchical statistical model that provides a more comprehensive representation of the benchmarking process by incorporating both benchmark characteristics and LLM randomness. We show that leveraging multiple generations improves the accuracy of estimating the benchmark score and reduces variance. We also introduce $\mathbb P\left(\text{correct}\right)$, a prompt-level difficulty score based on correct ratios, providing fine-grained insights into individual prompts. Additionally, we create a data map that visualizes difficulty and semantic prompts, enabling error detection and quality control in benchmark construction.

Transformer-based language models have achieved significant success; however, their internal mechanisms remain largely opaque due to the complexity of non-linear interactions and high-dimensional operations. While previous studies have demonstrated that these models implicitly embed reasoning trees, humans typically employ various distinct logical reasoning mechanisms to complete the same task. It is still unclear which multi-step reasoning mechanisms are used by language models to solve such tasks. In this paper, we aim to address this question by investigating the mechanistic interpretability of language models, particularly in the context of multi-step reasoning tasks. Specifically, we employ circuit analysis and self-influence functions to evaluate the changing importance of each token throughout the reasoning process, allowing us to map the reasoning paths adopted by the model. We apply this methodology to the GPT-2 model on a prediction task (IOI) and demonstrate that the underlying circuits reveal a human-interpretable reasoning process used by the model.

This study explored how Vision-Language Models (VLMs) process ignorance implicatures with visual and linguistic cues. Particularly, we focused on the effects of contexts (precise and approximate contexts) and modifier types (bare numerals, superlative, and comparative modifiers), which were considered pragmatic and semantic factors respectively. Methodologically, we conducted a truth-value judgment task in visually grounded settings using GPT-4o and Gemini 1.5 Pro. The results indicate that while both models exhibited sensitivity to linguistic cues (modifier), they failed to process ignorance implicatures with visual cues (context) as humans do. Specifically, the influence of context was weaker and inconsistent across models, indicating challenges in pragmatic reasoning for VLMs. On the other hand, superlative modifiers were more strongly associated with ignorance implicatures as compared to comparative modifiers, supporting the semantic view. These findings highlight the need for further advancements in VLMs to process language-vision information in a context-dependent way to achieve human-like pragmatic inference.

Inductive Reasoning (IR), the ability to summarize rules from examples and apply on new ones, has long been viewed as a primal ability for general intelligence and widely studied by cognitive science and AI researchers. Many benchmarks have been proposed to measure such ability for Large Language Models (LLMs); however, they focus on few-shot (usually $<$10) setting and lack evaluation for aggregating many pieces of information from long contexts. On the other hand, the ever-growing context length of LLMs have brought forth the novel paradigm of many-shot In-Context Learning (ICL), which addresses new tasks with hundreds to thousands of examples without expensive and inefficient fine-tuning. However, many-shot evaluations are mostly focused on classification (a very limited aspect of IR), and popular long-context LLM tasks such as Needle-In-A-Haystack (NIAH) seldom require complicated intelligence for integrating many pieces of information. To fix the issues from both worlds, we propose MIR-Bench, the first many-shot in-context inductive reasoning benchmark that asks LLM to induce output via input-output examples from underlying functions with diverse data format. Based on MIR-Bench, we study many novel problems for inductive reasoning and many-shot ICL, including robustness against erroneous shots and the effect of Chain-of-Thought (CoT), and acquired insightful findings.

With the rise of multimodal large language models, GPT-4o stands out as a pioneering model, driving us to evaluate its capabilities. This report assesses GPT-4o across various tasks to analyze its audio processing and reasoning abilities. We find that GPT-4o exhibits strong knowledge in audio, speech, and music understanding, performing well in tasks like intent classification, spoken command classification, semantic and grammatical reasoning., multilingual speech recognition, and singing analysis. It also shows greater robustness against hallucinations than other large audio-language models (LALMs). However, it struggles with tasks such as audio duration prediction and instrument classification. Additionally, GPT-4o's safety mechanisms cause it to decline tasks like speaker identification, age classification, MOS prediction, and audio deepfake detection. Notably, the model exhibits a significantly different refusal rate when responding to speaker verification tasks on different datasets. This is likely due to variations in the accompanying instructions or the quality of the input audio, suggesting the sensitivity of its built-in safeguards. Finally, we acknowledge that model performance varies with evaluation protocols. This report only serves as a preliminary exploration of the current state of LALMs.

Reasoning LLMs such as OpenAI o1, o3 and DeepSeek R1 have made significant progress in mathematics and coding, yet find challenging advanced tasks such as International Mathematical Olympiad (IMO) combinatorics problems, Abstraction and Reasoning Corpus (ARC) puzzles, and Humanity's Last Exam (HLE) questions. We use a diverse inference approach that combines multiple models and methods at test time. We find that verifying mathematics and code problems, and rejection sampling on other problems is simple and effective. We automatically verify correctness of solutions to IMO problems by Lean, and ARC puzzles by code, and find that best-of-N effectively answers HLE questions. Our approach increases answer accuracy on IMO combinatorics problems from 33.3% to 77.8%, accuracy on HLE questions from 8% to 37%, and solves 80% of ARC puzzles that 948 humans could not and 26.5% of ARC puzzles that o3 high compute does not. Test-time simulations, reinforcement learning, and meta-learning with inference feedback improve generalization by adapting agent graph representations and varying prompts, code, and datasets. Our approach is reliable, robust, and scalable, and in the spirit of reproducible research, we will make it publicly available upon publication.