# CHEAR: Confusion-Hierarchy Evidence-Aware Reasoning for Cost-Efficient Human Activity Recognition

arXiv: https://arxiv.org/abs/2610.15298 · HTML: https://mlsanity.com/paper/2610.15298/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning.
- Builds on 17 of its 31 references that are in the map.
- First posted on arXiv 11 Oct 2026.

## Placed in

- LLM reasoning / [Knowledge- and tool-augmented reasoning](https://mlsanity.com/lr/knowledge-and-tool-augmented-reasoning/)

## Builds on (most cited first, 17 of 17)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [ImageBind: One Embedding Space To Bind Them All](https://mlsanity.com/paper/2305.05665/) (2023)
- [DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models](https://mlsanity.com/paper/2512.02556/) (2025)
- [MOMENT: A Family of Open Time-series Foundation Models](https://mlsanity.com/paper/2402.03885/) (2024)
- [AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data](https://mlsanity.com/paper/2003.06505/) (2020)
- [The Faiss library](https://mlsanity.com/paper/2401.08281/) (2024)
- [Scaling Wearable Foundation Models](https://mlsanity.com/paper/2410.13638/) (2024)
- [HARGPT: Are LLMs Zero-Shot Human Activity Recognizers?](https://mlsanity.com/paper/2403.02727/) (2024)
- [SensorLLM: Aligning Large Language Models with Motion Sensors for Human Activity Recognition](https://mlsanity.com/paper/2410.10624/) (2024)
- [IMU2CLIP: Multimodal Contrastive Learning for IMU Motion Sensors from Egocentric Videos and Text](https://mlsanity.com/paper/2210.14395/) (2022)
- [Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals](https://mlsanity.com/paper/2412.09758/) (2024)
- [UniMTS: Unified Pre-training for Motion Time Series](https://mlsanity.com/paper/2410.19818/) (2024)
- [Mantis: Lightweight Foundation Model for Time Series Classification](https://mlsanity.com/paper/2502.15637/) (2025)
- [Generating Virtual On-body Accelerometer Data from Virtual Textual Descriptions for Human Activity Recognition](https://mlsanity.com/paper/2305.03187/) (2023)
- [ZARA: Training-Free Motion Time-Series Reasoning via Evidence-Grounded LLM Agents](https://mlsanity.com/paper/2508.04038/) (2025)
- [RAG-HAR: Retrieval Augmented Generation-based Human Activity Recognition](https://mlsanity.com/paper/2512.08984/) (2025)
- [RAG-HAR+: Towards Cost-Efficient LLM-Based Human Activity Recognition for Edge Deployment](https://mlsanity.com/paper/2607.26631/) (2026)

## Abstract (arXiv)

Large language models (LLMs) offer a new approach to wearable-sensor-based human activity recognition. Retrieval-augmented generation (RAG) further introduces similar motion examples and discriminative sensor knowledge, enabling decisions grounded in observations relevant to the current query. More evidence, however, does not necessarily mean better-organized evidence. When online prompts and LLM agents primarily orchestrate inference, decisions about which activities to compare and when to invoke discriminative knowledge remain concentrated at inference time. As the activity vocabulary grows, the model must repeatedly process more candidates and associated knowledge, whereas pruning candidates too early risks excluding the true activity. We propose confusion-hierarchy evidence-aware reasoning (CHEAR), which moves part of this knowledge organization offline. CHEAR constructs and validates an activity hierarchy from recurrent cross-subject retrieval errors and builds corresponding activity-pair knowledge. At inference time, the hierarchy orders candidate comparisons; the LLM retains candidates progressively, and local discriminative knowledge is accessed only for the remaining activities. Across eight datasets, CHEAR achieves 86.79% mean cross-subject accuracy, a 75.91% candidate reduction rate, and 96.44% true-activity retention. Per-sample token use is 85.15% lower on average than in a variant without the hierarchical structure. Mean accuracy across eight cross-dataset transfer directions is 88.81%. These results show that organizing candidates and discriminative knowledge around observed activity confusions reduces online LLM input while maintaining recognition performance.