Valerie Porter Shailesh Manjunath ((better)) -

In a 2024 study (Porter et al., ACM Transactions on Computer-Human Interaction ), she demonstrated that users rated an ATM-enabled chatbot as 37% more trustworthy than a baseline model, even when the baseline correctly identified real-time emotions. Her conclusion: consistency across interactions matters more than single-instance accuracy.

Manjunath’s critics argue that sparse sampling misses subtle affective shifts over minutes-long conversations. He has responded by developing adaptive sampling rates, but the trade-off between efficiency and emotional granularity remains unresolved. 4. Comparative Analysis and Synthesis | Dimension | Valerie Porter | Shailesh Manjunath | |-----------|----------------|---------------------| | Temporal focus | Long-term affective memory | Real-time, momentary inference | | Primary modality | Conversational history + user modeling | Multimodal (face, voice, text) | | Hardware requirement | Moderate (cloud or hybrid) | Low (edge-only, privacy-preserving) | | Key strength | Trust and relational coherence | Speed, privacy, scalability | | Key weakness | High storage; risk of affective bias | May miss gradual emotional change | valerie porter shailesh manjunath

Critics note that ATM requires significant storage and computational overhead. Porter herself acknowledged that long-term affective traces risk reinforcing negative stereotypes (e.g., persistently treating a user as “angry” after one outburst). This opens the door to Manjunath’s engineering solutions. 3. Shailesh Manjunath: Real-Time Multimodal Affect Processing 3.1 Key Contributions Manjunath’s work, presented at ICML 2023 and IEEE Affective Computing 2025, focuses on lightweight transformer models that fuse facial micro-expressions, vocal prosody, and keystroke dynamics. His signature system, AffectEdge , runs entirely on-device, addressing privacy concerns inherent in cloud-based emotion recognition. In a 2024 study (Porter et al

Unlike Porter’s longitudinal approach, Manjunath prioritizes sparse temporal sampling —analyzing only 3–5 emotion-relevant frames per second rather than continuous video. In a 2025 field deployment for a teletherapy platform, AffectEdge achieved 89% accuracy in detecting user distress while reducing latency to 120ms (vs. 2.3s for cloud models). He has responded by developing adaptive sampling rates,