Solving complex mathematical problems via system-2 reasoning is a natural human skill, yet it remains a significant challenge for current large language models (LLMs). We identify the scarcity of deliberate multi-step reasoning data as a primary limiting factor. To this end, we introduce Enriched Instruction Tuning (EIT), a method that enriches existing human-annotated mathematical datasets by synergizing human and AI feedback to create fine-grained reasoning trajectories. These datasets are then used to fine-tune open-source LLMs, enhancing their mathematical reasoning abilities without reliance on any symbolic verification program. Concretely, EIT is composed of two critical steps: Enriching with Reasoning Plan (ERP) and Enriching with Reasoning Step (ERS). The former generates a high-level plan that breaks down complex instructions into a sequence of simpler objectives, while ERS fills in reasoning contexts often overlooked by human annotators, creating a smoother reasoning trajectory for LLM fine-tuning. Unlike existing CoT prompting methods that generate reasoning chains only depending on LLM's internal knowledge, our method leverages human-annotated initial answers as ``meta-knowledge'' to help LLMs generate more detailed and precise reasoning processes, leading to a more trustworthy LLM expert for complex mathematical problems. In experiments, EIT achieves an accuracy of 84.1% on GSM8K and 32.5% on MATH, surpassing state-of-the-art fine-tuning and prompting methods, and even matching the performance of tool-augmented methods.

Growing labor shortages are increasing the demand for domestic service robots (DSRs) to assist in various settings. In this study, we develop a DSR that transports everyday objects to specified pieces of furniture based on open-vocabulary instructions. Our approach focuses on retrieving images of target objects and receptacles from pre-collected images of indoor environments. For example, given an instruction "Please get the right red towel hanging on the metal towel rack and put it in the white washing machine on the left," the DSR is expected to carry the red towel to the washing machine based on the retrieved images. This is challenging because the correct images should be retrieved from thousands of collected images, which may include many images of similar towels and appliances. To address this, we propose RelaX-Former, which learns diverse and robust representations from among positive, unlabeled positive, and negative samples. We evaluated RelaX-Former on a dataset containing real-world indoor images and human annotated instructions including complex referring expressions. The experimental results demonstrate that RelaX-Former outperformed existing baseline models across standard image retrieval metrics. Moreover, we performed physical experiments using a DSR to evaluate the performance of our approach in a zero-shot transfer setting. The experiments involved the DSR to carry objects to specific receptacles based on open-vocabulary instructions, achieving an overall success rate of 75%.

Transformer-based large language models (LLMs) use the key-value (KV) cache to significantly accelerate inference by storing the key and value embeddings of past tokens. However, this cache consumes significant GPU memory. In this work, we introduce HashEvict, an algorithm that uses locality-sensitive hashing (LSH) to compress the KV cache. HashEvict quickly locates tokens in the cache that are cosine dissimilar to the current query token. This is achieved by computing the Hamming distance between binarized Gaussian projections of the current token query and cached token keys, with a projection length much smaller than the embedding dimension. We maintain a lightweight binary structure in GPU memory to facilitate these calculations. Unlike existing compression strategies that compute attention to determine token retention, HashEvict makes these decisions pre-attention, thereby reducing computational costs. Additionally, HashEvict is dynamic - at every decoding step, the key and value of the current token replace the embeddings of a token expected to produce the lowest attention score. We demonstrate that HashEvict can compress the KV cache by 30%-70% while maintaining high performance across reasoning, multiple-choice, long-context retrieval and summarization tasks.

Large language models (LLMs) are capable of solving a wide range of tasks, yet they have struggled with reasoning. To address this, we propose $\textbf{Additional Logic Training (ALT)}$, which aims to enhance LLMs' reasoning capabilities by program-generated logical reasoning samples. We first establish principles for designing high-quality samples by integrating symbolic logic theory and previous empirical insights. Then, based on these principles, we construct a synthetic corpus named $\textbf{Formal Logic Deduction Diverse}$ ($\textbf{FLD}$$_{\times 2}$), comprising numerous samples of multi-step deduction with unknown facts, diverse reasoning rules, diverse linguistic expressions, and challenging distractors. Finally, we empirically show that ALT on FLD$_{\times2}$ substantially enhances the reasoning capabilities of state-of-the-art LLMs, including LLaMA-3.1-70B. Improvements include gains of up to 30 points on logical reasoning benchmarks, up to 10 points on math and coding benchmarks, and 5 points on the benchmark suite BBH.

The rapid development of large language models (LLMs) necessitates robust, unbiased, and scalable methods for evaluating their capabilities. However, human annotations are expensive to scale, model-based evaluations are prone to biases in answer style, while target-answer-based benchmarks are vulnerable to data contamination and cheating. To address these limitations, we propose StructTest, a novel benchmark that evaluates LLMs on their ability to produce compositionally specified structured outputs as an unbiased, cheap-to-run and difficult-to-cheat measure. The evaluation is done deterministically by a rule-based evaluator, which can be easily extended to new tasks. By testing structured outputs across diverse task domains -- including Summarization, Code, HTML and Math -- we demonstrate that StructTest serves as a good proxy for general reasoning abilities, as producing structured outputs often requires internal logical reasoning. We believe that StructTest offers a critical, complementary approach to objective and robust model evaluation.

Large Language Models (LLMs) have demonstrated exceptional abilities across a broad range of language-related tasks, including generating solutions to complex reasoning problems. An effective technique to enhance LLM performance is in-context learning, which encourages a step-by-step reasoning process by including explanatory examples to guide the model's responses. However, selecting appropriate exemplars for the model poses a challenge, as each dataset demands a distinct set of exemplars to enable the LLM to learn effectively and perform well on the test set. Current studies often rely on uncertainty- or diversity-based selection strategies to select exemplars for annotation and to improve model learning. However, these studies typically employ a non-adaptive approach, selecting a set of exemplars all at once. We argue that this non-adaptive strategy may result in a set of exemplars with high redundancy in terms of the knowledge covered, ultimately reducing their overall informativeness. To address this limitation, we propose \textsc{Adaptive-Prompt}, a novel method that adaptively selects exemplars by leveraging model feedback from previously chosen exemplars. Experimental results show that \textsc{Adaptive-Prompt} significantly enhances LLM performance across a variety of reasoning tasks.

At present, all textual backdoor attack methods are based on single triggers: for example, inserting specific content into the text to activate the backdoor; or changing the abstract text features. The former is easier to be identified by existing defense strategies due to its obvious characteristics; the latter, although improved in invisibility, has certain shortcomings in terms of attack performance, construction of poisoned datasets, and selection of the final poisoning rate. On this basis, this paper innovatively proposes a Dual-Trigger backdoor attack based on syntax and mood, and optimizes the construction of the poisoned dataset and the selection strategy of the final poisoning rate. A large number of experimental results show that this method significantly outperforms the previous methods based on abstract features in attack performance, and achieves comparable attack performance (almost 100% attack success rate) with the insertion-based method. In addition, the two trigger mechanisms included in this method can be activated independently in the application phase of the model, which not only improves the flexibility of the trigger style, but also enhances its robustness against defense strategies. These results profoundly reveal that textual backdoor attacks are extremely harmful and provide a new perspective for security protection in this field.

Anxiety and depression are the most common mental health issues worldwide, affecting a non-negligible part of the population. Accordingly, stakeholders, including governments' health systems, are developing new strategies to promote early detection and prevention from a holistic perspective (i.e., addressing several disorders simultaneously). In this work, an entirely novel system for the multi-label classification of anxiety and depression is proposed. The input data consists of dialogues from user interactions with an assistant chatbot. Another relevant contribution lies in using Large Language Models (llms) for feature extraction, provided the complexity and variability of language. The combination of llms, given their high capability for language understanding, and Machine Learning (ml) models, provided their contextual knowledge about the classification problem thanks to the labeled data, constitute a promising approach towards mental health assessment. To promote the solution's trustworthiness, reliability, and accountability, explainability descriptions of the model's decision are provided in a graphical dashboard. Experimental results on a real dataset attain 90 % accuracy, improving those in the prior literature. The ultimate objective is to contribute in an accessible and scalable way before formal treatment occurs in the healthcare systems.

Training and deploying machine learning models relies on a large amount of human-annotated data. As human labeling becomes increasingly expensive and time-consuming, recent research has developed multiple strategies to speed up annotation and reduce costs and human workload: generating synthetic training data, active learning, and hybrid labeling. This tutorial is oriented toward practical applications: we will present the basics of each strategy, highlight their benefits and limitations, and discuss in detail real-life case studies. Additionally, we will walk through best practices for managing human annotators and controlling the quality of the final dataset. The tutorial includes a hands-on workshop, where attendees will be guided in implementing a hybrid annotation setup. This tutorial is designed for NLP practitioners from both research and industry backgrounds who are involved in or interested in optimizing data labeling projects.

In the domain of Multimodal Large Language Models (MLLMs), achieving human-centric video understanding remains a formidable challenge. Existing benchmarks primarily emphasize object and action recognition, often neglecting the intricate nuances of human emotions, behaviors, and speech visual alignment within video content. We present HumanVBench, an innovative benchmark meticulously crafted to bridge these gaps in the evaluation of video MLLMs. HumanVBench comprises 17 carefully designed tasks that explore two primary dimensions: inner emotion and outer manifestations, spanning static and dynamic, basic and complex, as well as single-modal and cross-modal aspects. With two advanced automated pipelines for video annotation and distractor-included QA generation, HumanVBench utilizes diverse state-of-the-art (SOTA) techniques to streamline benchmark data synthesis and quality assessment, minimizing human annotation dependency tailored to human-centric multimodal attributes. A comprehensive evaluation across 16 SOTA video MLLMs reveals notable limitations in current performance, especially in cross-modal and temporal alignment, underscoring the necessity for further refinement toward achieving more human-like understanding. HumanVBench is open-sourced to facilitate future advancements and real-world applications in video MLLMs.