Online Session #1 of ICCE-Berlin 2026: Emerging Topics and Solutions in CE
Agenda:
- "Beyond Photorealism: Interaction Intelligence Shapes Consumer Trust in XR Virtual Sales Agents"
- "TA-SAINT for Chronic Metabolic Disease Risk Prediction Using Temporal-Aware Contrastive Learning"
- "An Embedded Voice Assistant for Conversational Memory Support"
- Chapter Meeting time
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
Loading virtual attendance info...
- Anton-Wilhelm-Amo-Straße 30
- Berlin, Berlin
- Germany 10117
- Building: Hilton Hotel
- Room Number: Corinth
Speakers
Yaojie Li
Beyond Photorealism: Consumer Perceptions Across XR Virtual Sales Agent Paradigms
Extended Reality (XR) virtual sales-agent systems combine visual embodiment and conversational capability through deployment paradigms with different practical trade-offs. This paper compares three bundled paradigms in a counterbalanced within-subjects experiment (N=24) situated in a simulated high-stakes immersive e-commerce scenario: (I) a high-fidelity non-interactive presenter; (II) an embodied conversational Wizard-of-Oz agent; and (III) a minimally embodied conversational agent powered by artificial intelligence. Paradigms I and II received higher form realism ratings, and following the post hoc decomposition of the behavioral realism measure, they received higher scores on the resulting physical behavioral realism than Paradigm III. Perceived interactivity and overall perceived trustworthiness increased from I to II to III. Behavioral intention was higher for III and II than for I. Eeriness did not significantly differ across paradigms. The findings characterize complete configurations rather than isolated components. They further motivate a three-dimensional conceptual framework and emphasize the importance of context. These insights provide context-bounded design implications for developing trustworthy XR virtual sales agents in immersive e-commerce.
Qingfeng Lu
TA-SAINT for Chronic Metabolic Disease Risk Prediction Using Temporal-Aware Contrastive Learning
Chronic metabolic diseases exhibit a protracted and insidious progression, making early dynamic risk warning a cross-disciplinary challenge. Longitudinal health examination data, the primary data source for this task, inherently present three critical challenges: short observation sequences, irregular sampling intervals, and heterogeneous feature spaces. Existing approaches often overlook the temporal dynamics embedded in such data and fail to provide a unified modeling paradigm. To address these limitations, we present TA-SAINT, a temporal-aware contrastive learning framework for chronic metabolic disease risk prediction. TA-SAINT explicitly models irregular temporal patterns while accommodating the properties of short-sequence health data, enabling a shift from static disease detection to dynamic risk forecasting. It employs dual attention to capture feature-level clinical interactions and population-level phenotypic similarities, together with temporal-weighted contrastive learning to account for uneven follow-up intervals. We evaluate TA-SAINT on a single-center longitudinal clinical health examination dataset and the China Health and Retirement Longitudinal Study cohort. Experiments across multiple metabolic risk prediction tasks show that TA-SAINT consistently outperforms conventional baseline models, validating its effectiveness and generalizability for irregular short-sequence medical data.
Ayaan Qayyum
An Embedded Voice Assistant for Conversational Memory Support
We propose an embedded device intended to support individuals with memory impairment by providing real time speaker identification. The system captures speech from a user’s environment and generates unique voice embeddings using a pipeline of Speech-To-Text and Speaker Reverification models. When a new speaker is detected, a unique identity is created and updated continuously via a running mean embedding. When a known speaker is detected, the system informs the user who is speaking and provides context from prior conversations. Initial evaluation on healthy speakers establishes technical feasibility, showing that audio windows of 5 seconds or longer achieve a near 100% identity match rate, while the embedding update is robust across a wide range of smoothing coefficients. Clinical validation with dementia patients remains future work. This prototype demonstrates that portable edge devices can provide persistent conversational awareness for memory impaired populations.
Agenda
The webinar will take place on the sidelines of the 15th IEEE International Conference on Consumer Electronics (ICCE Berlin 2026), held on 5–7 September 2026 in Berlin, Germany.
Greetings and Introduction: 5 minutes
Presentations: 15-20 minutes (each)
Q&A: 3-5 minutes (each)
Chapter time: 15-30 minutes