DEV Community

Papers Mache profile picture

Papers Mache

404 bio not found

Joined Joined on 
AI/ML Research Digest — Nov 14, 2026

AI/ML Research Digest — Nov 14, 2026

Comments
4 min read
Flow matching replaces brittle reconstruction pipelines

Flow matching replaces brittle reconstruction pipelines

Comments
2 min read
SIMPLE law predicts quadratic prefill cost

SIMPLE law predicts quadratic prefill cost

Comments
3 min read
Branch‑parallel draft distillation speeds text generation 3.6

Branch‑parallel draft distillation speeds text generation 3.6

Comments
2 min read
Sub‑bit KV cache quantization doubles throughput

Sub‑bit KV cache quantization doubles throughput

Comments
2 min read
Single-step flow matching brightens low‑light photos

Single-step flow matching brightens low‑light photos

Comments
2 min read
Speculative decoding cuts speech latency below 250 ms

Speculative decoding cuts speech latency below 250 ms

Comments
2 min read
TailSteer reduces drift in multi‑turn image editing; History Forcing stabilizes long‑horizon video generation

TailSteer reduces drift in multi‑turn image editing; History Forcing stabilizes long‑horizon video generation

Comments
2 min read
AI/ML Research Digest — Nov 07, 2026

AI/ML Research Digest — Nov 07, 2026

Comments
5 min read
Information‑gain design trims training compute

Information‑gain design trims training compute

Comments
2 min read
Tiny per‑option encoders beat larger joints

Tiny per‑option encoders beat larger joints

Comments
2 min read
Reference‑free benchmarks expose early‑stop overconfidence

Reference‑free benchmarks expose early‑stop overconfidence

Comments
2 min read
Calibration‑free top‑k MoE selector saves tuning

Calibration‑free top‑k MoE selector saves tuning

Comments
2 min read
Quantcheck flags unsafe low‑bit LLMs

Quantcheck flags unsafe low‑bit LLMs

Comments
2 min read
Trope memory powers coherent long‑form stories

Trope memory powers coherent long‑form stories

Comments
2 min read
Latent 3D Gaussian memory enables real‑time rendering

Latent 3D Gaussian memory enables real‑time rendering

Comments
2 min read
AI/ML Research Digest — Oct 31, 2026

AI/ML Research Digest — Oct 31, 2026

Comments
4 min read
Synthetic high‑value questions boost distillation efficiency

Synthetic high‑value questions boost distillation efficiency

Comments
2 min read
On‑policy distillation works with limited additional data

On‑policy distillation works with limited additional data

Comments
2 min read
Multimodal benchmark exposes 37% evidence gap

Multimodal benchmark exposes 37% evidence gap

Comments
2 min read
Memory‑native model shrinks token history size ~66

Memory‑native model shrinks token history size ~66

Comments
2 min read
Adaptive compute reduces latency by up to ~45%

Adaptive compute reduces latency by up to ~45%

Comments
3 min read
Iterating layers at inference lifts reasoning scores

Iterating layers at inference lifts reasoning scores

Comments
2 min read
AI/ML Research Digest — Oct 24, 2026

AI/ML Research Digest — Oct 24, 2026

Comments
5 min read
Box regression head boosts visual grounding speed, accuracy

Box regression head boosts visual grounding speed, accuracy

Comments
2 min read
Stateless agents cut long‑horizon token usage

Stateless agents cut long‑horizon token usage

Comments
2 min read
Plug‑and‑play harness evolution boosts benchmarks without weight changes

Plug‑and‑play harness evolution boosts benchmarks without weight changes

Comments
2 min read
Video diffusion compute drops from thousands to one hour

Video diffusion compute drops from thousands to one hour

Comments
2 min read
INT4 caching doubles long‑context attention throughput

INT4 caching doubles long‑context attention throughput

Comments
2 min read
Reward shaping slashes reasoning steps by 65%

Reward shaping slashes reasoning steps by 65%

Comments
2 min read
AI/ML Research Digest — Oct 17, 2026

AI/ML Research Digest — Oct 17, 2026

Comments
6 min read
Label‑free token uncertainty improves confidence

Label‑free token uncertainty improves confidence

Comments
2 min read
Accelerator certification speeds VLA models 10

Accelerator certification speeds VLA models 10

Comments
2 min read
Shared neural renderer cuts agent learning rounds

Shared neural renderer cuts agent learning rounds

Comments
2 min read
Expert pruning improves decode throughput up to 1.8 with minimal quality loss

Expert pruning improves decode throughput up to 1.8 with minimal quality loss

Comments
2 min read
Self‑distilled spatial reasoning improves accuracy without large latency penalty

Self‑distilled spatial reasoning improves accuracy without large latency penalty

Comments
2 min read
Evidence Ledger Boosts Multimodal Reasoning by Up to 15%

Evidence Ledger Boosts Multimodal Reasoning by Up to 15%

Comments
2 min read
Token‑level routing can achieve up to 64 higher decoding throughput

Token‑level routing can achieve up to 64 higher decoding throughput

Comments
2 min read
Multimodal indexes can recreate images

Multimodal indexes can recreate images

Comments
2 min read
Object permanence dataset powers new reasoning record

Object permanence dataset powers new reasoning record

Comments
2 min read
64 Visual Token Compression Preserves Accuracy

64 Visual Token Compression Preserves Accuracy

Comments
2 min read
Soft Memory Extends Context Length, Gives 4 Speedups

Soft Memory Extends Context Length, Gives 4 Speedups

Comments
2 min read
Shared Fisher‑Rao geometry enables universal control

Shared Fisher‑Rao geometry enables universal control

Comments
1 min read
AI/ML Research Digest — Oct 10, 2026

AI/ML Research Digest — Oct 10, 2026

Comments
5 min read
2-Bit KV Cache Slashes Memory Bandwidth

2-Bit KV Cache Slashes Memory Bandwidth

Comments
2 min read
Code‑derived skill bank lifts benchmarks 12%

Code‑derived skill bank lifts benchmarks 12%

Comments 1
2 min read
Early-layer multimodal bias is detectable

Early-layer multimodal bias is detectable

Comments
2 min read
Octree Video Memory Enables Minute-Long Consistency

Octree Video Memory Enables Minute-Long Consistency

Comments
2 min read
Episodic memory boosts visual agents sevenfold

Episodic memory boosts visual agents sevenfold

Comments 1
2 min read
One-Step Diffusion Refiner Cuts Latency Ninefold

One-Step Diffusion Refiner Cuts Latency Ninefold

Comments 1
2 min read
Realtime‑Venus sets state‑of‑the‑art multimodal scores

Realtime‑Venus sets state‑of‑the‑art multimodal scores

Comments
2 min read
5% Tokens, 14% Accuracy Gain via Self-Distillation

5% Tokens, 14% Accuracy Gain via Self-Distillation

Comments
2 min read
AI/ML Research Digest — Sep 27, 2026

AI/ML Research Digest — Sep 27, 2026

Comments
3 min read
AI/ML Research Digest — Oct 04, 2026

AI/ML Research Digest — Oct 04, 2026

Comments
4 min read
Anchors boost memorization twenty‑eightfold

Anchors boost memorization twenty‑eightfold

1
Comments
2 min read
KV cache cut by ~45% with near‑same accuracy

KV cache cut by ~45% with near‑same accuracy

Comments
2 min read
Open‑vocab relation predictor triples recall

Open‑vocab relation predictor triples recall

1
Comments
2 min read
Recursive self‑improvement cuts token traffic 49%

Recursive self‑improvement cuts token traffic 49%

Comments
2 min read
35 B MoE runs under 3 GiB RAM

35 B MoE runs under 3 GiB RAM

Comments
2 min read
AI/ML Research Digest — Sep 19, 2026

AI/ML Research Digest — Sep 19, 2026

Comments
4 min read
loading...