Fed-HeLLo: Efficient Federated Foundation Model Fine-Tuning with Heterogeneous LoRA Allocation

Federated Learning has recently been utilized to collaboratively fine-tune foundation models across multiple clients. Notably, federated low-rank adaptation LoRA-based fine-tuning methods have recently gained attention, which allows clients to fine-tune FMs with a small portion of trainable parameters locally. However, most existing methods do not account for the heterogeneous resources of clients or lack an effective local training strategy to maximize global fine-tuning performance under limited resources. In this work, we propose Fed-HeLLo, a novel federated LoRA-based fine-tuning framework that enables clients to collaboratively fine-tune an FM with different local trainable LoRA layers. To ensure its effectiveness, we develop several heterogeneous LoRA allocation (HLA) strategies that adaptively allocate local trainable LoRA layers based on clients' resource capabilities and the layer importance. Specifically, based on the dynamic layer importance, we design a Fisher Information Matrix score-based HLA that leverages dynamic gradient norm information. To better stabilize the training process, we consider the intrinsic importance of LoRA layers and design a Geometrically-Defined HLA strategy. It shapes the collective distribution of trainable LoRA layers into specific geometric patterns, such as Triangle, Inverted Triangle, Bottleneck, and Uniform. Moreover, we extend GD-HLA into a randomized version, named Randomized Geometrically-Defined HLA, for enhanced model accuracy with randomness. By co-designing the proposed HLA strategies, we incorporate both the dynamic and intrinsic layer importance into the design of our HLA strategy. We evaluate our approach on five datasets under diverse federated LoRA fine-tuning settings, covering three levels of data distribution from IID to extreme Non-IID. Results show that Fed-HeLLo with HLA strategies is both effective and efficient.

Private GPTs for LLM-driven testing in software development and machine learning

In this contribution, we examine the capability of private GPTs to automatically generate executable test code based on requirements. More specifically, we use acceptance criteria as input, formulated as part of epics, or stories, which are typically used in modern development processes. This gives product owners, or business intelligence, respectively, a way to directly produce testable criteria through the use of LLMs. We explore the quality of the so-produced tests in two ways: i) directly by letting the LLM generate code from requirements, ii) through an intermediate step using Gherkin syntax. As a result, it turns out that the two-step procedure yields better results -where we define better in terms of human readability and best coding practices, i.e. lines of code and use of additional libraries typically used in testing. Concretely, we evaluate prompt effectiveness across two scenarios: a simple "Hello World" program and a digit classification model, showing that structured prompts lead to higher-quality test outputs.

'Hello, World!': Making GNNs Talk with LLMs

While graph neural networks (GNNs) have shown remarkable performance across diverse graph-related tasks, their high-dimensional hidden representations render them black boxes. In this work, we propose Graph Lingual Network (GLN), a GNN built on large language models (LLMs), with hidden representations in the form of human-readable text. Through careful prompt design, GLN incorporates not only the message passing module of GNNs but also advanced GNN techniques, including graph attention and initial residual connection. The comprehensibility of GLN's hidden representations enables an intuitive analysis of how node representations change (1) across layers and (2) under advanced GNN techniques, shedding light on the inner workings of GNNs. Furthermore, we demonstrate that GLN achieves strong zero-shot performance on node classification and link prediction, outperforming existing LLM-based baseline methods.

The AI Gap: How Socioeconomic Status Affects Language Technology Interactions

Socioeconomic status (SES) fundamentally influences how people interact with each other and more recently, with digital technologies like Large Language Models (LLMs). While previous research has highlighted the interaction between SES and language technology, it was limited by reliance on proxy metrics and synthetic data. We survey 1,000 individuals from diverse socioeconomic backgrounds about their use of language technologies and generative AI, and collect 6,482 prompts from their previous interactions with LLMs. We find systematic differences across SES groups in language technology usage (i.e., frequency, performed tasks), interaction styles, and topics. Higher SES entails a higher level of abstraction, convey requests more concisely, and topics like 'inclusivity' and 'travel'. Lower SES correlates with higher anthropomorphization of LLMs (using ''hello'' and ''thank you'') and more concrete language. Our findings suggest that while generative language technologies are becoming more accessible to everyone, socioeconomic linguistic differences still stratify their use to exacerbate the digital divide. These differences underscore the importance of considering SES in developing language technologies to accommodate varying linguistic needs rooted in socioeconomic factors and limit the AI Gap across SES groups.

Universal Embedding Function for Traffic Classification via QUIC Domain Recognition Pretraining: A Transfer Learning Success

Encrypted traffic classification (TC) methods must adapt to new protocols and extensions as well as to advancements in other machine learning fields. In this paper, we follow a transfer learning setup best known from computer vision. We first pretrain an embedding model on a complex task with a large number of classes and then transfer it to five well-known TC datasets. The pretraining task is recognition of SNI domains in encrypted QUIC traffic, which in itself is a problem for network monitoring due to the growing adoption of TLS Encrypted Client Hello. Our training pipeline -- featuring a disjoint class setup, ArcFace loss function, and a modern deep learning architecture -- aims to produce universal embeddings applicable across tasks. The proposed solution, based on nearest neighbors search in the embedding space, surpasses SOTA performance on four of the five TC datasets. A comparison with a baseline method utilizing raw packet sequences revealed unexpected findings with potential implications for the broader TC field. We published the model architecture, trained weights, and transfer learning experiments.

Implicit Communication in Human-Robot Collaborative Transport

We focus on human-robot collaborative transport, in which a robot and a user collaboratively move an object to a goal pose. In the absence of explicit communication, this problem is challenging because it demands tight implicit coordination between two heterogeneous agents, who have very different sensing, actuation, and reasoning capabilities. Our key insight is that the two agents can coordinate fluently by encoding subtle, communicative signals into actions that affect the state of the transported object. To this end, we design an inference mechanism that probabilistically maps observations of joint actions executed by the two agents to a set of joint strategies of workspace traversal. Based on this mechanism, we define a cost representing the human's uncertainty over the unfolding traversal strategy and introduce it into a model predictive controller that balances between uncertainty minimization and efficiency maximization. We deploy our framework on a mobile manipulator (Hello Robot Stretch) and evaluate it in a within-subjects lab study (N=24). We show that our framework enables greater team performance and empowers the robot to be perceived as a significantly more fluent and competent partner compared to baselines lacking a communicative mechanism.

Siamese Networks for Cat Re-Identification: Exploring Neural Models for Cat Instance Recognition

Street cats in urban areas often rely on human intervention for survival, leading to challenges in population control and welfare management. In April 2023, Hello Inc., a Chinese urban mobility company, launched the Hello Street Cat initiative to address these issues. The project deployed over 21,000 smart feeding stations across 14 cities in China, integrating livestreaming cameras and treat dispensers activated through user donations. It also promotes the Trap-Neuter-Return (TNR) method, supported by a community-driven platform, HelloStreetCatWiki, where volunteers catalog and identify cats. However, manual identification is inefficient and unsustainable, creating a need for automated solutions. This study explores Deep Learning-based models for re-identifying street cats in the Hello Street Cat initiative. A dataset of 2,796 images of 69 cats was used to train Siamese Networks with EfficientNetB0, MobileNet and VGG16 as base models, evaluated under contrastive and triplet loss functions. VGG16 paired with contrastive loss emerged as the most effective configuration, achieving up to 97% accuracy and an F1 score of 0.9344 during testing. The approach leverages image augmentation and dataset refinement to overcome challenges posed by limited data and diverse visual variations. These findings underscore the potential of automated cat re-identification to streamline population monitoring and welfare efforts. By reducing reliance on manual processes, the method offers a scalable and reliable solution for communitydriven initiatives. Future research will focus on expanding datasets and developing real-time implementations to enhance practicality in large-scale deployments.

ANAVI: Audio Noise Awareness using Visuals of Indoor environments for NAVIgation

We propose Audio Noise Awareness using Visuals of Indoors for NAVIgation for quieter robot path planning. While humans are naturally aware of the noise they make and its impact on those around them, robots currently lack this awareness. A key challenge in achieving audio awareness for robots is estimating how loud will the robot's actions be at a listener's location? Since sound depends upon the geometry and material composition of rooms, we train the robot to passively perceive loudness using visual observations of indoor environments. To this end, we generate data on how loud an 'impulse' sounds at different listener locations in simulated homes, and train our Acoustic Noise Predictor (ANP). Next, we collect acoustic profiles corresponding to different actions for navigation. Unifying ANP with action acoustics, we demonstrate experiments with wheeled (Hello Robot Stretch) and legged (Unitree Go2) robots so that these robots adhere to the noise constraints of the environment. See code and data at https://anavi-corl24.github.io/

Coherence-Driven Multimodal Safety Dialogue with Active Learning for Embodied Agents

When assisting people in daily tasks, robots need to accurately interpret visual cues and respond effectively in diverse safety-critical situations, such as sharp objects on the floor. In this context, we present M-CoDAL, a multimodal-dialogue system specifically designed for embodied agents to better understand and communicate in safety-critical situations. The system leverages discourse coherence relations to enhance its contextual understanding and communication abilities. To train this system, we introduce a novel clustering-based active learning mechanism that utilizes an external Large Language Model (LLM) to identify informative instances. Our approach is evaluated using a newly created multimodal dataset comprising 1K safety violations extracted from 2K Reddit images. These violations are annotated using a Large Multimodal Model (LMM) and verified by human annotators. Results with this dataset demonstrate that our approach improves resolution of safety situations, user sentiment, as well as safety of the conversation. Next, we deploy our dialogue system on a Hello Robot Stretch robot and conduct a within-subject user study with real-world participants. In the study, participants role-play two safety scenarios with different levels of severity with the robot and receive interventions from our model and a baseline system powered by OpenAI's ChatGPT. The study results corroborate and extend the findings from automated evaluation, showing that our proposed system is more persuasive and competent in a real-world embodied agent setting.

Denial-of-Service Poisoning Attacks against Large Language Models

Recent studies have shown that LLMs are vulnerable to denial-of-service (DoS) attacks, where adversarial inputs like spelling errors or non-semantic prompts trigger endless outputs without generating an [EOS] token. These attacks can potentially cause high latency and make LLM services inaccessible to other users or tasks. However, when there are speech-to-text interfaces (e.g., voice commands to a robot), executing such DoS attacks becomes challenging, as it is difficult to introduce spelling errors or non-semantic prompts through speech. A simple DoS attack in these scenarios would be to instruct the model to "Keep repeating Hello", but we observe that relying solely on natural instructions limits output length, which is bounded by the maximum length of the LLM's supervised finetuning (SFT) data. To overcome this limitation, we propose poisoning-based DoS (P-DoS) attacks for LLMs, demonstrating that injecting a single poisoned sample designed for DoS purposes can break the output length limit. For example, a poisoned sample can successfully attack GPT-4o and GPT-4o mini (via OpenAI's finetuning API) using less than $1, causing repeated outputs up to the maximum inference length (16K tokens, compared to 0.5K before poisoning). Additionally, we perform comprehensive ablation studies on open-source LLMs and extend our method to LLM agents, where attackers can control both the finetuning dataset and algorithm. Our findings underscore the urgent need for defenses against P-DoS attacks to secure LLMs. Our code is available at https://github.com/sail-sg/P-DoS.