Early-stage ML researcher advancing post-training ternary quantization of LLMs, from Qwen3-4B to Qwen3-8B
3
papers here
2
reviewed by Evolve
3
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Output metrics are not yet scorable: OpenAlex provides no reliable total-publication or venue record for this identity beyond three 2026 preprints. This axis would measure publication volume, output consistency across years, and authorship roles.
Impact—
Citation impact cannot be scored: the only available signals are h-index 0 and i10-index 0, reflecting 2026 arXiv preprints too recent for meaningful citation accumulation. This axis would measure citation counts, h-index trajectory, and broader scholarly influence over time.
Focus0/100
The focus score of 0 reflects a still-consolidating research identity: OpenAlex topics scatter across four areas — Topic Modeling and Advanced Neural Network Applications (3 works each), Low-power high-performance VLSI design (2), and Domain Adaptation and Few-Shot Learning (1) — while the field distribution splits 0.5/0.5 between two near-duplicate efficient-inference and model-compression labels. The three catalog papers themselves are tightly clustered on ternary quantization.
Collaboration22/100
Collaboration scores 22, consistent with a compact, small-team mode: OpenAlex lists only 2 distinct coauthors, and the frequent-collaborator list names M Sparsh Mehra and Poojith Devan. Malik is first author on all three 2026 papers (none as corresponding author), indicating a lead-contributor role within a two-person working pattern rather than a large consortium.
Activity40/100
Activity scores 40 on a short but current record: OpenAlex counts 3 works in the recent three-year window, all in 2026. The burst includes two reviewed compression papers and one degradation study, indicating active ongoing effort on ternary quantization, but with no prior-year record the trajectory cannot yet be judged as sustained.
Reproducibility55/100
Reproducibility scores 55, averaged over 2 reviewed papers, both rated medium repro-readiness. Evolve's verdicts credit unusually candid reporting — including a rare reproduction gate in the Qwen3-8B study — while flagging weaknesses that temper confidence: single-seed n=500 evaluations, absent competing baselines, and a 7.9% FP16 gap. Transparency is a stated strength; verification breadth is not yet.
Research summary
Anirudh Malik is an early-stage machine learning researcher whose recent work centers on post-training quantization, specifically ternary (1.58-bit) weight representations for large language models. All three papers attributed to this identity in the Evolve catalog are 2026 arXiv preprints with Malik as first author: a study ternarizing Qwen3-4B to 1.641 bits per weight (accuracy down 9.8 points, storage from 8.29 to 3.96 GiB), a scale-up to Qwen3-8B covering capability retention, reproduction, lossless packing, and packed execution, and a third paper on capability-stratified degradation in ternary language models. Evolve's reviews credit candid, transparent reporting — including a rare reproduction gate — while flagging single-seed n=500 evaluations, missing competing baselines, and a 7.9% FP16 gap. Bibliometric signals are correspondingly early: OpenAlex reports h-index 0 and i10-index 0, three works in the recent three-year window, and two distinct coauthors, M Sparsh Mehra and Poojith Devan. OpenAlex topics reach beyond compression into topic modeling, neural network applications, and low-power VLSI design, though the field distribution is split evenly across two efficient-inference and model-compression labels. No institution, country, or ORCID is on record, so this profile is name-resolved from public bibliographic data.
Collaboration pattern
Malik's collaboration footprint is small and stable: OpenAlex records 2 distinct coauthors, and the frequent-collaborator list names M Sparsh Mehra and Poojith Devan. On all three 2026 catalog papers Malik is first author and never corresponding author, indicating a consistent lead-author role within a compact team. No cross-institution or senior-collaborator pattern can be inferred, since affiliation and correspondence details are absent from the record.
Research keywords
Highlights
Scaling ternarisation to Qwen3-8B
Qwen3-4B ternarized to 1.641 bits per weight
Capability-stratified degradation in ternary LMs
A third 2026 first-author arXiv paper analyzing how degradation stratifies by capability in ternary language models; not yet reviewed in the Evolve catalog, so no verdict or repro-readiness is on record.
Timeline
- 2026First cataloged paper: post-training ternarization of Qwen3-4B, reaching 1.641 bits/weight with storage cut from 8.29 to 3.96 GiB.
- 2026Scale-up to Qwen3-8B spanning capability retention, reproduction, lossless packing, and packed execution; Evolve notes a rare reproduction gate.
- 2026Third first-author arXiv paper on capability-stratified degradation in ternary language models completes a three-paper 2026 series.
- 2026OpenAlex records all 3 works within 2026 (h-index 0, i10-index 0, 2 distinct coauthors) — an active but very early citation profile.
Papers in the Evolve catalog
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.
- Identity is name-resolved from OpenAlex bibliographic data; no institution, country, or ORCID is on record, and attribution carries some uncertainty for this name.
- All bibliometrics come from an OpenAlex snapshot; the output and impact axes are marked missing and were not scored.
- Scope covers only three 2026 arXiv preprints with no citation counts or venue history yet — an early-stage profile.
- Evolve's qualitative verdicts cover two of the three papers; the capability-stratified degradation paper has no review on record.