Harish Haresamudram

Harish Haresamudram, Irfan Essa, Thomas Ploetz

The dichotomy between the challenging nature of obtaining annotations for activities, and the more straightforward nature of data collection from wearables, has resulted in significant interest in the development of techniques that utilize large quantities of unlabeled data for learning representations. Contrastive Predictive Coding (CPC) is one such method, learning effective representations by leveraging properties of time-series data to setup a contrastive future timestep prediction task. In this work, we propose enhancements to CPC, by systematically investigating the encoder architecture, the aggregator network, and the future timestep prediction, resulting in a fully convolutional architecture, thereby improving parallelizability. Across sensor positions and activities, our method shows substantial improvements on four of six target datasets, demonstrating its ability to empower a wide range of application scenarios. Further, in the presence of very limited labeled data, our technique significantly outperforms both supervised and self-supervised baselines, positively impacting situations where collecting only a few seconds of labeled data may be possible. This is promising, as CPC does not require specialized data transformations or reconstructions for learning effective representations.

Harish Haresamudram, Apoorva Beedu, Mashfiqui Rabbi et al.

Cross-modal contrastive pre-training between natural language and other modalities, e.g., vision and audio, has demonstrated astonishing performance and effectiveness across a diverse variety of tasks and domains. In this paper, we investigate whether such natural language supervision can be used for wearable sensor based Human Activity Recognition (HAR), and discover that-surprisingly-it performs substantially worse than standard end-to-end training and self-supervision. We identify the primary causes for this as: sensor heterogeneity and the lack of rich, diverse text descriptions of activities. To mitigate their impact, we also develop strategies and assess their effectiveness through an extensive experimental evaluation. These strategies lead to significant increases in activity recognition, bringing performance closer to supervised and self-supervised training, while also enabling the recognition of unseen activities and cross modal retrieval of videos. Overall, our work paves the way for better sensor-language learning, ultimately leading to the development of foundational models for HAR using wearables.

Harish Haresamudram, Irfan Essa, Thomas Ploetz

Human activity recognition (HAR) in wearable computing is typically based on direct processing of sensor data. Sensor readings are translated into representations, either derived through dedicated preprocessing, or integrated into end-to-end learning. Independent of their origin, for the vast majority of contemporary HAR, those representations are typically continuous in nature. That has not always been the case. In the early days of HAR, discretization approaches have been explored - primarily motivated by the desire to minimize computational requirements, but also with a view on applications beyond mere recognition, such as, activity discovery, fingerprinting, or large-scale search. Those traditional discretization approaches, however, suffer from substantial loss in precision and resolution in the resulting representations with detrimental effects on downstream tasks. Times have changed and in this paper we propose a return to discretized representations. We adopt and apply recent advancements in Vector Quantization (VQ) to wearables applications, which enables us to directly learn a mapping between short spans of sensor data and a codebook of vectors, resulting in recognition performance that is generally on par with their contemporary, continuous counterparts - sometimes surpassing them. Therefore, this work presents a proof-of-concept for demonstrating how effective discrete representations can be derived, enabling applications beyond mere activity classification but also opening up the field to advanced tools for the analysis of symbolic sequences, as they are known, for example, from domains such as natural language processing. Based on an extensive experimental evaluation on a suite of wearables-based benchmark HAR tasks, we demonstrate the potential of our learned discretization scheme and discuss how discretized sensor data analysis can lead to substantial changes in HAR.

Hyeongju Choi, Apoorva Beedu, Harish Haresamudram et al.

To properly assist humans in their needs, human activity recognition (HAR) systems need the ability to fuse information from multiple modalities. Our hypothesis is that multimodal sensors, visual and non-visual tend to provide complementary information, addressing the limitations of other modalities. In this work, we propose a multi-modal framework that learns to effectively combine features from RGB Video and IMU sensors, and show its robustness for MMAct and UTD-MHAD datasets. Our model is trained in two-stage, where in the first stage, each input encoder learns to effectively extract features, and in the second stage, learns to combine these individual features. We show significant improvements of 22% and 11% compared to video only and IMU only setup on UTD-MHAD dataset, and 20% and 12% on MMAct datasets. Through extensive experimentation, we show the robustness of our model on zero shot setting, and limited annotated data setting. We further compare with state-of-the-art methods that use more input modalities and show that our method outperforms significantly on the more difficult MMact dataset, and performs comparably in UTD-MHAD dataset.

Megha Thukral, Harish Haresamudram, Thomas Ploetz

The ubiquitous availability of smartphones and smartwatches with integrated inertial measurement units (IMUs) enables straightforward capturing of human activities. For specific applications of sensor based human activity recognition (HAR), however, logistical challenges and burgeoning costs render especially the ground truth annotation of such data a difficult endeavor, resulting in limited scale and diversity of datasets. Transfer learning, i.e., leveraging publicly available labeled datasets to first learn useful representations that can then be fine-tuned using limited amounts of labeled data from a target domain, can alleviate some of the performance issues of contemporary HAR systems. Yet they can fail when the differences between source and target conditions are too large and/ or only few samples from a target application domain are available, each of which are typical challenges in real-world human activity recognition scenarios. In this paper, we present an approach for economic use of publicly available labeled HAR datasets for effective transfer learning. We introduce a novel transfer learning framework, Cross-Domain HAR, which follows the teacher-student self-training paradigm to more effectively recognize activities with very limited label information. It bridges conceptual gaps between source and target domains, including sensor locations and type of activities. Through our extensive experimental evaluation on a range of benchmark datasets, we demonstrate the effectiveness of our approach for practically relevant few shot activity recognition scenarios. We also present a detailed analysis into how the individual components of our framework affect downstream performance.

Richard Goldman, Varun Komperla, Thomas Ploetz et al.

A promising alternative to the computationally expensive Neural Architecture Search (NAS) involves the development of Zero Cost Proxies (ZCPs), which correlate well with trained performance, but can be computed through a single forward/backward pass on a randomly sampled batch of data. In this paper, we investigate the effectiveness of ZCPs for HAR on six benchmark datasets, and demonstrate that they discover network architectures that obtain within 5% of performance attained by full-scale training involving 1500 randomly sampled architectures. This results in substantial computational savings as high-performing architectures can be discovered with minimal training. Our experiments not only introduce ZCPs to sensor-based HAR, but also demonstrate that they are robust to data noise, further showcasing their suitability for practical scenarios.

Megha Thukral, Sourish Gunesh Dhekane, Shruthi K. Hiremath et al.

Human activity recognition (HAR) using ambient sensors in smart homes has numerous applications for human healthcare and wellness. However, building general-purpose HAR models that can be deployed to new smart home environments requires a significant amount of annotated sensor data and training overhead. Most smart homes vary significantly in their layouts, i.e., floor plans and the specifics of sensors embedded, resulting in low generalizability of HAR models trained for specific homes. We address this limitation by introducing a novel, layout-agnostic modeling approach for HAR systems in smart homes that utilizes the transferrable representational capacity of natural language descriptions of raw sensor data. To this end, we generate Textual Descriptions Of Sensor Triggers (TDOST) that encapsulate the surrounding trigger conditions and provide cues for underlying activities to the activity recognition models. Leveraging textual embeddings, rather than raw sensor data, we create activity recognition systems that predict standard activities across homes without either (re-)training or adaptation on target homes. Through an extensive evaluation, we demonstrate the effectiveness of TDOST-based models in unseen smart homes through experiments on benchmarked CASAS datasets. Furthermore, we conduct a detailed analysis of how the individual components of our approach affect downstream activity recognition performance.

Harish Haresamudram, Chi Ian Tang, Sungho Suh et al.

In the many years since the inception of wearable sensor-based Human Activity Recognition (HAR), a wide variety of methods have been introduced and evaluated for their ability to recognize activities. Substantial gains have been made since the days of hand-crafting heuristics as features, yet, progress has seemingly stalled on many popular benchmarks, with performance falling short of what may be considered 'sufficient'-- despite the increase in computational power and scale of sensor data, as well as rising complexity in techniques being employed. The HAR community approaches a new paradigm shift, this time incorporating world knowledge from foundational models. In this paper, we take stock of sensor-based HAR -- surveying it from its beginnings to the current state of the field, and charting its future. This is accompanied by a hands-on tutorial, through which we guide practitioners in developing HAR systems for real-world application scenarios. We provide a compendium for novices and experts alike, of methods that aim at finally solving the activity recognition problem.

Apoorva Beedu, Harish Haresamudram, Karan Samel et al.

Anticipating future actions is a highly challenging task due to the diversity and scale of potential future actions; yet, information from different modalities help narrow down plausible action choices. Each modality can provide diverse and often complementary context for the model to learn from. While previous multi-modal methods leverage information from modalities such as video and audio, we primarily explore how text descriptions of actions and objects can also lead to more accurate action anticipation by providing additional contextual cues, e.g., about the environment and its contents. We propose a Multi-modal Contrastive Anticipative Transformer (M-CAT), a video transformer architecture that jointly learns from multi-modal features and text descriptions of actions and objects. We train our model in two stages, where the model first learns to align video clips with descriptions of future actions, and is subsequently fine-tuned to predict future actions. Compared to existing methods, M-CAT has the advantage of learning additional context from two types of text inputs: rich descriptions of future actions during pre-training, and, text descriptions for detected objects and actions during modality feature fusion. Through extensive experimental evaluation, we demonstrate that our model outperforms previous methods on the EpicKitchens datasets, and show that using simple text descriptions of actions and objects aid in more effective action anticipation. In addition, we examine the impact of object and action information obtained via text, and perform extensive ablations.

Wanting Mao, Maxwell A Xu, Harish Haresamudram et al.

Modeling multi-modal time-series data is critical for capturing system-level dynamics, particularly in biosignals where modalities such as ECG, PPG, EDA, and accelerometry provide complementary perspectives on interconnected physiological processes. While recent self-supervised learning (SSL) advances have improved unimodal representation learning, existing multi-modal approaches often rely on CLIP-style contrastive objectives that overfit to easily aligned features and misclassify valid cross-modal relationships as negatives, resulting in fragmented and non-generalizable embeddings. To overcome these limitations, we propose ProtoMM, a novel SSL framework that introduces a shared prototype dictionary to anchor heterogeneous modalities in a common embedding space. By clustering representations around shared prototypes rather than explicit negative sampling, our method captures complementary information across modalities and provides a coherent "common language" for physiological signals. In this work, we focus on developing a Pulse Motion foundation model with ProtoMM and demonstrate that our approach outperforms contrastive-only and prior multimodal SSL methods, achieving state-of-the-art performance while offering improved interpretability of learned features.

Harish Haresamudram, Hrudhai Rajasekhar, Nikhil Murlidhar Shanbhogue et al.

The astonishing success of Large Language Models (LLMs) in Natural Language Processing (NLP) has spurred their use in many application domains beyond text analysis, including wearable sensor-based Human Activity Recognition (HAR). In such scenarios, often sensor data are directly fed into an LLM along with text instructions for the model to perform activity classification. Seemingly remarkable results have been reported for such LLM-based HAR systems when they are evaluated on standard benchmarks from the field. Yet, we argue, care has to be taken when evaluating LLM-based HAR systems in such a traditional way. Most contemporary LLMs are trained on virtually the entire (accessible) internet -- potentially including standard HAR datasets. With that, it is not unlikely that LLMs actually had access to the test data used in such benchmark experiments.The resulting contamination of training data would render these experimental evaluations meaningless. In this paper we investigate whether LLMs indeed have had access to standard HAR datasets during training. We apply memorization tests to LLMs, which involves instructing the models to extend given snippets of data. When comparing the LLM-generated output to the original data we found a non-negligible amount of matches which suggests that the LLM under investigation seems to indeed have seen wearable sensor data from the benchmark datasets during training. For the Daphnet dataset in particular, GPT-4 is able to reproduce blocks of sensor readings. We report on our investigations and discuss potential implications on HAR research, especially with regards to reporting results on experimental evaluation

Harish Haresamudram, Irfan Essa, Thomas Plötz

The emergence of self-supervised learning in the field of wearables-based human activity recognition (HAR) has opened up opportunities to tackle the most pressing challenges in the field, namely to exploit unlabeled data to derive reliable recognition systems for scenarios where only small amounts of labeled training samples can be collected. As such, self-supervision, i.e., the paradigm of 'pretrain-then-finetune' has the potential to become a strong alternative to the predominant end-to-end training approaches, let alone hand-crafted features for the classic activity recognition chain. Recently a number of contributions have been made that introduced self-supervised learning into the field of HAR, including, Multi-task self-supervision, Masked Reconstruction, CPC, and SimCLR, to name but a few. With the initial success of these methods, the time has come for a systematic inventory and analysis of the potential self-supervised learning has for the field. This paper provides exactly that. We assess the progress of self-supervised HAR research by introducing a framework that performs a multi-faceted exploration of model performance. We organize the framework into three dimensions, each containing three constituent criteria, such that each dimension captures specific aspects of performance, including the robustness to differing source and target conditions, the influence of dataset characteristics, and the feature space characteristics. We utilize this framework to assess seven state-of-the-art self-supervised methods for HAR, leading to the formulation of insights into the properties of these techniques and to establish their value towards learning representations for diverse scenarios.

Harish Haresamudram, Irfan Essa, Thomas Ploetz

Feature extraction is crucial for human activity recognition (HAR) using body-worn movement sensors. Recently, learned representations have been used successfully, offering promising alternatives to manually engineered features. Our work focuses on effective use of small amounts of labeled data and the opportunistic exploitation of unlabeled data that are straightforward to collect in mobile and ubiquitous computing scenarios. We hypothesize and demonstrate that explicitly considering the temporality of sensor data at representation level plays an important role for effective HAR in challenging scenarios. We introduce the Contrastive Predictive Coding (CPC) framework to human activity recognition, which captures the long-term temporal structure of sensor data streams. Through a range of experimental evaluations on real-life recognition tasks, we demonstrate its effectiveness for improved HAR. CPC-based pre-training is self-supervised, and the resulting learned representations can be integrated into standard activity chains. It leads to significantly improved recognition performance when only small amounts of labeled training data are available, thereby demonstrating the practical value of our approach.

Hyeokhyen Kwon, Catherine Tong, Harish Haresamudram et al.

The lack of large-scale, labeled data sets impedes progress in developing robust and generalized predictive models for on-body sensor-based human activity recognition (HAR). Labeled data in human activity recognition is scarce and hard to come by, as sensor data collection is expensive, and the annotation is time-consuming and error-prone. To address this problem, we introduce IMUTube, an automated processing pipeline that integrates existing computer vision and signal processing techniques to convert videos of human activity into virtual streams of IMU data. These virtual IMU streams represent accelerometry at a wide variety of locations on the human body. We show how the virtually-generated IMU data improves the performance of a variety of models on known HAR datasets. Our initial results are very promising, but the greater promise of this work lies in a collective approach by the computer vision, signal processing, and activity recognition communities to extend this work in ways that we outline. This should lead to on-body, sensor-based HAR becoming yet another success story in large-dataset breakthroughs in recognition.