Stevens Institute of Technology researcher in streaming speech, multimodal continual learning, and LLM privacy.
4
papers here
4
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not scored: the underlying record reports h-index 0, i10-index 0, and no publication history before the 8 works logged in 2026, so cumulative productivity cannot be measured. This axis would count total and per-year outputs once a stable author record accumulates.
Impact—
Not scored: with h-index 0 and i10-index 0, the record carries no citation counts, so influence cannot be measured yet. This axis would track citations and uptake of the 2026 arXiv preprints as citation data accrues.
Focus0/100
Focus scores 0, the maximal-spread signal in this dataset: field distribution splits evenly at 0.25 across four areas — speech processing for TTS, speech processing for ASR, machine learning for continual and multimodal learning, and AI security and privacy for NLP security — while five OpenAlex topics carry 2 works each. The work is genuinely multi-front; streaming inference is the nearest thing to a unifying theme.
Collaboration63/100
Collaboration scores 63: the record lists 23 distinct co-authors, with a wide, non-dominated pool of frequent collaborators including Zehan Liu, Rime Wen, Kaiqi Fu, Shawn Qin, Roy Gan, Carl Chen, Altman Lin, Lights Shi, Qian Wang, Kai Guo, Chuanbin Liu, and Peng Hu. All four reviewed papers are team efforts — three middle-authored, one last-authored — and no corresponding authorship is recorded, suggesting a contributing rather than lab-leading role so far.
Activity64/100
Activity scores 64 on a compressed but real signal: OpenAlex logs 8 works in 2026, the only year with counts and the entirety of the recent three-year total (recent3yWorks 8). All four reviewed items are arXiv preprints released that year. The cadence is recent and burst-like rather than sustained, so the score reflects current momentum more than a long-run publishing habit.
Reproducibility46/100
Reproducibility scores 46, averaging 4 reviewed papers (reproMean 46) rated medium, low, medium, medium in readiness. Evolve's checks flag recurring gaps: teacher-relative gains on synthetic speech for the TTS tracker, open comparability questions for the GRPO-trained streaming ASR, no seed variance for NeuCME, and a single-run attack with no released code for the embedding-obfuscation study. Results look promising but are not yet independently verifiable.
Research summary
Hao Wang is a researcher at Stevens Institute of Technology whose 2026 output spans four equally weighted directions: streaming text-to-speech, streaming automatic speech recognition, multimodal continual learning, and attacks on privacy-preserving LLM inference. The OpenAlex snapshot records 8 works in 2026, the only year with activity, while h-index and i10-index stand at 0 — a very early-stage profile whose impact data has not accrued. Among the four papers Evolve has reviewed, all 2026 arXiv preprints, Wang is a middle author on three and last author on X2Streaming-ASR, a GRPO-trained wait/emit policy that cuts streaming recognition commit latency to 24-97 ms against 409-585 ms baselines. Further work includes X2-NativeCursor, a native-token observer for text progress in streaming codec TTS (MAE 0.151 vs 1.253); NeuCME, a claimed-first dynamic multimodal continual learning framework built on rehearsal, multi-gated mixture-of-experts, and relevance distillation; and PMA, an unsupervised Gromov-Wasserstein alignment attack that defeats deterministic embedding obfuscation. The record lists 23 distinct co-authors and a rotating pool of frequent collaborators, with no corresponding-author role recorded. Reproducibility averages 46 across the four reviewed papers, with reviewers citing single-run evidence, teacher-relative or narrow gains, and limited code release. Hao Wang is a common name; this profile is name-resolved from public bibliographic data.
Collaboration pattern
The record shows 23 distinct co-authors, with a wide, non-dominated pool of frequent collaborators: Zehan Liu, Rime Wen, Kaiqi Fu, Shawn Qin, Roy Gan, Carl Chen, Altman Lin, Lights Shi, Qian Wang, Kai Guo, Chuanbin Liu, and Peng Hu. All four reviewed papers are team efforts, mostly middle-authored, with one last-author placement on X2Streaming-ASR and no corresponding authorship recorded — a contributing rather than lab-leading pattern so far.
Research keywords
Highlights
Wait/emit policy cuts streaming ASR commit latency to 24-97 ms
Native-token observer tracks text progress in streaming codec TTS
Dynamic multimodal continual learning via neural combinatorics
Unsupervised manifold-alignment attack breaks LLM embedding obfuscation
Timeline
- 2026OpenAlex logs 8 works in this single year, the only year with counts and the entirety of the recent three-year output in the snapshot.
- 2026Co-authors PMA, an unsupervised Gromov-Wasserstein alignment attack against embedding obfuscation in privacy-preserving LLM inference.
- 2026Co-authors NeuCME, a claimed-first dynamic multimodal continual learning framework with a new benchmark.
- 2026Last author on X2Streaming-ASR, a GRPO-trained wait/emit streaming ASR system with 24-97 ms commit latency.
- 2026Co-authors X2-NativeCursor for native-token text progress tracking in streaming codec TTS.
Papers in the Evolve catalog
X2-NativeCursor: Native-Token Text Progress Tracking for Incremental-Text Streaming Codec TTS
X2Streaming-ASR: wait when uncertain, emit when ready for streaming ASR
NeuCME: Toward Dynamic Multimodal Continual Learning via Neural Combinatorics of Multiple Experts
A Novel Semantic Manifold Alignment Attack against Embedding-to-Embedding Obfuscation in Privacy-Preserving LLMs
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric indicators come from an OpenAlex author record only; no ORCID is linked, and Hao Wang is a common name, so this profile is name-resolved from public bibliographic data and attribution may shift.
- Output and impact axes are unscored: h-index and i10-index read 0 with no citation aggregates, reflecting a thin or fragmented record rather than measured influence.
- All four reviewed papers are 2026 arXiv preprints; peer-review status and final venues are not reflected.
- The reproducibility score (mean 46) averages 4 reviewed papers rated medium or low readiness; reviewers note single-run evidence, teacher-relative or narrow gains, and no released code in several cases.