Recent advances in large language models (LLMs) have significantly impacted the domain of multi-hop question answering (MHQA), where systems are required to aggregate information and infer answers from disparate pieces of text. However, the autoregressive nature of LLMs inherently poses a challenge as errors may accumulate if mistakes are made in the intermediate reasoning steps. This paper introduces Monte-Carlo tree search for Zero-shot multi-hop Question Answering (MZQA), a framework based on Monte-Carlo tree search (MCTS) to identify optimal reasoning paths in MHQA tasks, mitigating the error propagation from sequential reasoning processes. Unlike previous works, we propose a zero-shot prompting method, which relies solely on instructions without the support of hand-crafted few-shot examples that typically require domain expertise. We also introduce a behavioral cloning approach (MZQA-BC) trained on self-generated MCTS inference trajectories, achieving an over 10-fold increase in reasoning speed with bare compromise in performance. The efficacy of our method is validated on standard benchmarks such as HotpotQA, 2WikiMultihopQA, and MuSiQue, demonstrating that it outperforms existing frameworks.

We present Timber, the first white-box poisoning attack targeting decision trees. Timber is based on a greedy attack strategy leveraging sub-tree retraining to efficiently estimate the damage performed by poisoning a given training instance. The attack relies on a tree annotation procedure which enables sorting training instances so that they are processed in increasing order of computational cost of sub-tree retraining. This sorting yields a variant of Timber supporting an early stopping criterion designed to make poisoning attacks more efficient and feasible on larger datasets. We also discuss an extension of Timber to traditional random forest models, which is useful because decision trees are normally combined into ensembles to improve their predictive power. Our experimental evaluation on public datasets shows that our attacks outperform existing baselines in terms of effectiveness, efficiency or both. Moreover, we show that two representative defenses can mitigate the effect of our attacks, but fail at effectively thwarting them.

In this work, we present AfriHuBERT, an extension of mHuBERT-147, a state-of-the-art (SOTA) and compact self-supervised learning (SSL) model, originally pretrained on 147 languages. While mHuBERT-147 was pretrained on 16 African languages, we expand this to cover 39 African languages through continued pretraining on 6,500+ hours of speech data aggregated from diverse sources, including 23 newly added languages. We evaluate AfriHuBERT on two key speech tasks: Language Identification (LID) and Automatic Speech Recognition (ASR) using FLEURS dataset. Our results show a +4% F1 score improvement on average for LID and a -1.2% average Word Error Rate (WER) reduction for ASR. Further analysis shows that ASR models trained on AfriHuBERT exhibit improved cross-corpus generalization. Additionally, the analysis indicates that the FLEURS have data quality limitations that may affect their suitability for evaluating low-resource African languages, suggesting the need for better evaluation benchmarks for these languages.

This study evaluates the performance of large language models (LLMs) as medical agents in Portuguese, aiming to develop a reliable and relevant virtual assistant for healthcare professionals. The HealthCareMagic-100k-en and MedQuAD datasets, translated from English using GPT-3.5, were used to fine-tune the ChatBode-7B model using the PEFT-QLoRA method. The InternLM2 model, with initial training on medical data, presented the best overall performance, with high precision and adequacy in metrics such as accuracy, completeness and safety. However, DrBode models, derived from ChatBode, exhibited a phenomenon of catastrophic forgetting of acquired medical knowledge. Despite this, these models performed frequently or even better in aspects such as grammaticality and coherence. A significant challenge was low inter-rater agreement, highlighting the need for more robust assessment protocols. This work paves the way for future research, such as evaluating multilingual models specific to the medical field, improving the quality of training data, and developing more consistent evaluation methodologies for the medical field.

Cluster analysis plays a crucial role in various domains and applications, such as customer segmentation in marketing. These contexts often involve multimodal data, including both tabular and textual datasets, making it challenging to represent hidden patterns for obtaining meaningful clusters. This study introduces ERASMO, a framework designed to fine-tune a pretrained language model on textually encoded tabular data and generate embeddings from the fine-tuned model. ERASMO employs a textual converter to transform tabular data into a textual format, enabling the language model to process and understand the data more effectively. Additionally, ERASMO produces contextually rich and structurally representative embeddings through techniques such as random feature sequence shuffling and number verbalization. Extensive experimental evaluations were conducted using multiple datasets and baseline approaches. Our results demonstrate that ERASMO fully leverages the specific context of each tabular dataset, leading to more precise and nuanced embeddings for accurate clustering. This approach enhances clustering performance by capturing complex relationship patterns within diverse tabular data.

Specialist language models (LMs) focus on a specific task or domain on which they often outperform generalist LMs of the same size. However, the specialist data needed to pretrain these models is only available in limited amount for most tasks. In this work, we build specialist models from large generalist training sets instead. We adjust the training distribution of the generalist data with guidance from the limited domain-specific data. We explore several approaches, with clustered importance sampling standing out. This method clusters the generalist dataset and samples from these clusters based on their frequencies in the smaller specialist dataset. It is scalable, suitable for pretraining and continued pretraining, it works well in multi-task settings. Our findings demonstrate improvements across different domains in terms of language modeling perplexity and accuracy on multiple-choice question tasks. We also present ablation studies that examine the impact of dataset sizes, clustering configurations, and model sizes. Generalist language models (LMs) can address a wide variety of tasks, but this generality comes at a cost (Brown et al., 2020). It necessitates a large training set representative of all prospective tasks, as well as a large model to fit such a comprehensive dataset.

Addressing multiagent decision problems in AI, especially those involving collaborative or competitive agents acting concurrently in a partially observable and stochastic environment, remains a formidable challenge. While Interactive Dynamic Influence Diagrams~(I-DIDs) have offered a promising decision framework for such problems, they encounter limitations when the subject agent encounters unknown behaviors exhibited by other agents that are not explicitly modeled within the I-DID. This can lead to sub-optimal responses from the subject agent. In this paper, we propose a novel data-driven approach that utilizes an encoder-decoder architecture, particularly a variational autoencoder, to enhance I-DID solutions. By integrating a perplexity-based tree loss function into the optimization algorithm of the variational autoencoder, coupled with the advantages of Zig-Zag One-Hot encoding and decoding, we generate potential behaviors of other agents within the I-DID that are more likely to contain their true behaviors, even from limited interactions. This new approach enables the subject agent to respond more appropriately to unknown behaviors, thus improving its decision quality. We empirically demonstrate the effectiveness of the proposed approach in two well-established problem domains, highlighting its potential for handling multi-agent decision problems with unknown behaviors. This work is the first time of using neural networks based approaches to deal with the I-DID challenge in agent planning and learning problems.

Robotic manipulation in open-world settings requires not only task execution but also the ability to detect and learn from failures. While recent advances in vision-language models (VLMs) and large language models (LLMs) have improved robots' spatial reasoning and problem-solving abilities, they still struggle with failure recognition, limiting their real-world applicability. We introduce AHA, an open-source VLM designed to detect and reason about failures in robotic manipulation using natural language. By framing failure detection as a free-form reasoning task, AHA identifies failures and provides detailed, adaptable explanations across different robots, tasks, and environments. We fine-tuned AHA using FailGen, a scalable framework that generates the first large-scale dataset of robotic failure trajectories, the AHA dataset. FailGen achieves this by procedurally perturbing successful demonstrations from simulation. Despite being trained solely on the AHA dataset, AHA generalizes effectively to real-world failure datasets, robotic systems, and unseen tasks. It surpasses the second-best model (GPT-4o in-context learning) by 10.3% and exceeds the average performance of six compared models including five state-of-the-art VLMs by 35.3% across multiple metrics and datasets. We integrate AHA into three manipulation frameworks that utilize LLMs/VLMs for reinforcement learning, task and motion planning, and zero-shot trajectory generation. AHA's failure feedback enhances these policies' performances by refining dense reward functions, optimizing task planning, and improving sub-task verification, boosting task success rates by an average of 21.4% across all three tasks compared to GPT-4 models.

Social dynamics in close human interactions pose significant challenges for Human Mesh Estimation (HME), particularly due to the complexity of physical contacts and the scarcity of training data. Addressing these challenges, we introduce a novel data generation method that utilizes Large Vision Language Models (LVLMs) to annotate contact maps which guide test-time optimization to produce paired image and pseudo-ground truth meshes. This methodology not only alleviates the annotation burden but also enables the assembly of a comprehensive dataset specifically tailored for close interactions in HME. Our Ask Pose Unite (APU) dataset, comprising over 6.2k human mesh pairs in contact covering diverse interaction types, is curated from images depicting naturalistic person-to-person scenes. We empirically show that using our dataset to train a diffusion-based contact prior, used as guidance during optimization, improves mesh estimation on unseen interactions. Our work addresses longstanding challenges of data scarcity for close interactions in HME enhancing the field's capabilities of handling complex interaction scenarios.

Generative AI, in particular large transformer models, are increasingly driving HPC system design in science and industry. We analyze performance characteristics of such transformer models and discuss their sensitivity to the transformer type, parallelization strategy, and HPC system features (accelerators and interconnects). We utilize a performance model that allows us to explore this complex design space and highlight its key components. We find that different transformer types demand different parallelism and system characteristics at different training regimes. Large Language Models are performant with 3D parallelism and amplify network needs only at pre-training scales with reduced dependence on accelerator capacity and bandwidth. On the other hand, long-sequence transformers, representative of scientific foundation models, place a more uniform dependence on network and capacity with necessary 4D parallelism. Our analysis emphasizes the need for closer performance modeling of different transformer types keeping system features in mind and demonstrates a path towards this. Our code is available as open-source.