Every paper checked — nothing hidden
All papers
Every paper checked in the week of 12–18 Oct 2026 · Small and efficient models — nothing is hidden. Brief papers first.
All2,430In brief3More (on-topic)455Not relevant1,957Not assessed6Older27Could not confirm6Blogs with no paper16All blogs19
| Paper | Topic | Relevance | Feedback |
|---|---|---|---|
| Function-Space Temporal Self-Distillation for Policy Optimization The paper introduces a single-policy optimization framework combining privileged adaptation with stored historical next-token distributions for language model reasoning. Ranked #421Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Proposes a compute-efficient policy optimization method effective under a short training budget. | |
| Provable Transformer Context Compression for In-Context Reinforcement Learning The paper analyzes the minimum context size required for transformer models to perform in-context reinforcement learning without parameter updates. Ranked #422Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Establishes memory and context compression bounds for transformer RL. | |
| Empirical Boundaries of Reservoir Computing for Few-Shot Sequence Pattern Tasks The paper empirically analyzes echo state network variants on synthetic sequence-to-sequence pattern tasks. Ranked #423Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Investigates randomized recurrent networks as computationally efficient alternatives. | |
| Memory-Efficient Subspace Optimization with Residual Correction: From Momentum SGD to Muon This paper presents a memory-efficient subspace optimization framework using residual correction to stabilize language model training. Ranked #424Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Reduces optimizer memory footprint during large language model training via subspace optimization. | |
| ProtModernBERT: Compute-Efficient Protein Language Modeling This paper introduces an efficient encoder-only protein language model incorporating modern architectural updates to lower training costs. Ranked #425Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Introduces architectural refinements to significantly reduce pre-training FLOPs and compute cost. | |
| LLM-jp-4 Technical Report The paper details the development, training data criteria, and performance of the open Japanese sovereign models, LLM-jp-4. Ranked #426Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Presents efficient open-weight models including an 8B dense model and a 32B-A3B mixture-of-experts model. | |
| DTDD: Divergence-Triggered Dynamic Distillation for Reliable On-Policy Supervision The paper introduces a distillation framework that triggers teacher takeovers when student policies diverge during long-horizon reasoning. Ranked #427Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Focuses on policy distillation to train models efficiently with significantly fewer updates. | |
| FOLD: Position-Free Far Memory with Local Order for Long-Context Speculative Decoding This paper introduces FOLD, a training-free speculative decoding approach treating distant tokens as position-free far memory for long-context generation. Ranked #428Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Provides a training-free speculative decoding method to accelerate long-context inference and reduce latency. | |
| TreeSpark: Learning to Draft the Trees You Verify for Speculative Decoding This paper introduces TreeSpark, a framework training semi-autoregressive drafters to generate candidate trees aligned with speculative verification. Ranked #429Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Improves inference throughput and serving efficiency via tree-based speculative decoding. | |
| RM-OPD: On-Policy Distillation for Generative Reward Models This paper presents RM-OPD, a framework that trains domain-specialized reward models via on-policy distillation into a generalist model. Ranked #430Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Uses on-policy distillation to consolidate multi-domain teacher models into a single student. | |
| Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study The paper tests a heuristic to set the number of transformer attention heads using the intrinsic dimension of token representations. Ranked #431Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Proposes a method to reduce attention parameters by setting head count from intrinsic dimension. | |
| JetSpec++: Foresight Parallel Drafting with Self Refinements The paper introduces a speculative decoding system combining self-refinement and pipelined drafting to reduce wall-clock generation time. Ranked #432Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Presents a parallel speculative decoding method that substantially speeds up inference serving. | |
| Qwen-MT-Lumen at WMT2026: Instruction-Specialized Translation via Quality-Aware Training and On-Policy Distillation The paper presents a four-stage training pipeline combining preference optimization and distillation for instruction-following machine translation. Ranked #433Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Uses on-policy distillation from a larger model to train an efficient 14B translation model. | |
| DUALVIEW: Leveraging Broad and Output-Specific Evidence for LLM-based Multi-Output Text Regression The paper proposes a parameter-efficient method leveraging frozen model representations for multi-output text regression tasks across diverse benchmarks. Ranked #434Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Enables highly parameter-efficient regression allowing smaller models to beat larger ones. | |
| PA1-OPD: One-Step Prefix-Augmented On-Policy Distillation for Language Model Pretraining The paper proposes a prefix-augmented distillation method providing one-step student branch feedback during language model pretraining. Ranked #435Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Develops an on-policy knowledge distillation method for language model pretraining. | |
| AgriJev: Efficient Agricultural Candidate Scoring with Multi-Scale Adapted Models The paper presents adapted small language models designed to directly score agricultural candidate choices in a single forward pass. Ranked #436Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Adapts compact models (0.8B-9B) to halve scoring latency compared to generative answer models. | |
| Decoupled Marginal Sharpening for Training-Free Inference-Time Scaling This paper proposes a decoupled marginal sharpening method to improve inference-time reasoning in language models without weight updates or reward models. Ranked #437Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Proposes an inference-time reasoning method that improves efficiency without training external verifiers. | |
| Beyond Expert Rankings: Contrast Calibration for Mixed-Precision MoE Quantization This paper proposes a contrast-calibrated bit allocation method for mixed-precision quantization of mixture-of-experts language models. Ranked #438Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Introduces mixed-precision quantization for MoE models to reduce storage and inference costs. | |
| SpectrumQuant: Spectrum Alignment for LLM Quantization This paper proposes a closed-form covariance-aligned rotation method to improve post-training low-bit quantization in large language models. Ranked #439Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Introduces an orthogonal transform method for low-bit large language model quantization. | |
| CAST: History-Adaptive Spectral Clipping for LLM Pretraining Introduces an adaptive spectral gradient clipping method to stabilize and accelerate large language model pretraining. Ranked #440Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Improves pretraining efficiency and step convergence for large MoE models. | |
| EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Ranked #441Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Proposes training-free test-time computation techniques for model adaptation. | |
| CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Ranked #442Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Designs an architecture enabling local small language models to perform complex reasoning effectively. | |
| Wolof-Adapted Industry-Ready Open-Weight Language Models The paper presents open-weight language models adapted for the Wolof language using instruction tuning, preference alignment, and model merging. Ranked #443Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Develops and adapts small 0.6B to 14B open-weight models for local deployment. | |
| Optimal Hydra: Model Heterogeneous Federated Learning for LLMs using Data-Informed Hydra-Nets The paper proposes a federated learning framework that tailors sub-network sizes and architectures to heterogeneous client hardware and data. Ranked #444Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Adapts transformer sub-network architectures to resource-constrained client devices in federated learning. | |
| Value-Informed Schedule Trimming for Accelerated Sequential Monte Carlo Guidance in Discrete Diffusion The paper introduces a training-free schedule optimization method for accelerating sequential Monte Carlo guidance in discrete diffusion models. Ranked #445Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Accelerates test-time discrete diffusion inference by dynamically reducing sequential Monte Carlo guidance steps. | |
| Ranking and Partitioning of Attention Heads for Sparse and Continual Reinforcement Learning Fine-tuning The paper studies ranking and partitioning attention heads by gradient norm for sparse and continual reinforcement learning fine-tuning. Ranked #446Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Proposes sparse attention-head tuning to reduce training and adaptation overhead across tasks. | |
| Data Strategies for Large Language Models: A Unified Perspective on the Training Lifecycle Reviews and formalizes data selection, weighting, and scheduling policies across the pretraining and post-training lifecycle of language models. Ranked #447Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Analyzes compute-budget allocation and training data efficiency across model lifecycles. | |
| BitSparse: Compact Activation Storage for ReLU Feed-Forward Networks Presents a GPU format that losslessly compresses sparse activations to reduce memory usage during neural network training. Ranked #448Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Introduces a lossless activation compression format to reduce GPU memory during LLM training. | |
| Query-Efficient Rejection Sampling for Parallel Masked Diffusion Proposes a single-increment rejection sampling algorithm to correct token correlation bias efficiently in parallel masked diffusion. Ranked #449Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Improves parallel masked diffusion decoding efficiency via single-query rejection sampling. | |
| LCT: Losslessly Compressed Tensors with GPU friendly execution for Training Provides a GPU-accelerated Huffman compression framework for tensors to lower memory overhead during model training. Ranked #450Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Presents lossless GPU tensor compression to reduce memory footprint during model training. | |
| ETH: Data-Balanced Heterogeneous Pipeline Parallelism for Training Large Models This paper presents a pipeline parallelism training system that balances data workloads across heterogeneous hardware while reducing communication overhead. Ranked #451Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Proposes pipeline parallelism to improve memory and compute efficiency for training large models. | |
| Learning Lean Feedback for Efficient Proof-Candidate Evaluation on GPUs This paper trains GPU-based neural evaluators to predict Lean 4 compiler verdicts and feedback, accelerating proof-candidate filtering. Ranked #452Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Accelerates formal verification by using lightweight neural surrogates to speed up theorem prover evaluation. | |
| Compiling Multi-Threshold Operators for Quantized Non-Monotonic and Fine-Granular Activations in Custom-Tailored DNN Accelerators The paper proposes an ONNX compiler method to map quantized non-monotonic and fine-granular activations into multi-threshold operators for hardware accelerators. Ranked #453Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Compiles multi-threshold operators to enable hardware-efficient quantized activation functions. | |
| Sparse Federated Echo State Networks for Time-Series Classification via Smooth-L1 Regularization The paper presents a sparse federated Echo State Network using smooth L1 regularization for efficient edge time-series classification. Ranked #454Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Presents a sparse regularized network to reduce stored model size and inference cost. | |
| ShiftQAT: Quantization-Aware Training for the Numerical Contract of a Fixed-Function Edge CNN Accelerator The paper details a quantization-aware training workflow that trains neural networks to match the exact hardware constraints of edge accelerators. Ranked #455Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Develops quantization-aware training specifically to satisfy constraints of edge accelerators. | |
| Recursive Offline Distillation via Entropy Gain Sampling and Answer Mixing The paper presents an offline distillation method combining entropy-difference sampling and answer mixing to transfer reasoning abilities into smaller language models. Ranked #456Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Proposes an offline distillation method to transfer reasoning capabilities into small models. | |
| Fault Detection and Bit-Error Rate Diagnostics in Optical Wireless Links Using Edge-AI The authors implement an on-device machine learning diagnostic engine on a microcontroller for optical wireless communication links. Ranked #457Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Demonstrates an INT8 quantized TinyML neural network deployed on a microcontroller for edge processing. | |
| RoLaQ: Joint Rotation and Range Calibration for 2-Bit KV Caches in Multi-Head Latent Attention This paper presents a calibration method to compress multi-head latent attention key-value caches to two bits for efficient long-context language model inference. Ranked #458Preprint0/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Proposes rotation and range calibration for 2-bit KV cache compression to improve inference efficiency. | |
| SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models The paper presents an 8-bit weight and activation post-training quantization technique that migrates activation outliers to weights. Ranked #459New way to see a problemPeer-reviewed7/10 suggestedOlder paper (2022) Older paper (2022) — discussed this week | Small and efficient models via blog · 12 Oct, 16:38 | On-topic Introduces SmoothQuant for 8-bit weight and activation quantization to accelerate LLM inference. | |
| AlignQuant: Tile-Aligned Mixed-Precision Quantization for Efficient LLM Generation The paper introduces a post-training mixed-precision quantization method using GPU-compatible 2D weight tiles for efficient inference. Ranked #460New way to see a problemPreprint6/10 suggestedOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via blog · 12 Oct, 16:38 | On-topic Introduces tile-aligned mixed-precision quantization to accelerate LLM inference and reduce GPU memory. | |
| FadeMem: Biologically-Inspired Forgetting for Efficient Agent Memory Proposes an agent memory architecture that uses biologically inspired adaptive decay functions to selectively retain and forget context. Ranked #461New way to see a problemPeer-reviewed6/10 suggestedOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via ieee · 12 Oct, 16:30 | On-topic Focuses on reducing agent memory storage requirements by 45 percent through selective forgetting mechanisms. | |
| Comparative Evaluation of GGUF and GPTQ Quantization for Llama 3.1 Inference on x86 and ARM Architectures The paper compares GGUF and GPTQ post-training quantization methods for running Llama 3.1 on x86 and ARM edge hardware. Ranked #462Surprising resultPeer-reviewed6/10 suggestedOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via ieee · 13 Oct, 06:19 | On-topic Evaluates GGUF and GPTQ quantization techniques for efficient LLM inference on resource-constrained hardware. | |
| Euston: Efficient and User-Friendly Secure Transformer Inference with Non-Interactivity This paper proposes a non-interactive secure inference framework for Transformer models using fully homomorphic encryption to reduce user overhead. Ranked #463Surprising resultPeer-reviewed6/10 suggestedOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via ieee · 14 Oct, 06:18 | On-topic Addresses efficient inference serving for Transformer-based LLMs using homomorphic encryption. | |
| TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones The paper presents an open-source multimodal language model combining a small language backbone with vision encoders for efficient inference. Ranked #465Preprint0/10 suggestedOlder paper (2023) Older paper (2023) — discussed this week | Small and efficient models via blog · 12 Oct, 16:38 | On-topic Develops an efficient multimodal model using a 2.8B parameter backbone requiring only 8GB for inference. | |
| AVNet (Accelerator-Optimized Convolutional Network): A Hardware-Efficient CNN for FPGA-based Deployment using HLS4ML for PCB Defect Detection This paper presents AVNet, an optimized convolutional network designed for efficient FPGA deployment to classify circuit board defects. Ranked #466Peer-reviewed0/10 suggestedOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via ieee · 14 Oct, 06:18 | On-topic Focuses on quantization and efficient FPGA deployment for lightweight neural network inference. | |
| Distilling Diffusion Models to Efficient 3D LiDAR Scene Completion Proposes a distillation framework and structural loss to speed up diffusion-based 3D LiDAR scene completion for autonomous driving perception. Ranked #467Peer-reviewed0/10 suggestedOlder paper (2024) Older paper (2024) — discussed this week | Small and efficient models via ieee · 12 Oct, 16:30 | On-topic Proposes distillation to accelerate diffusion model sampling speed by over five times for efficient inference. | |
| LP-Spec: Leveraging LPDDR PIM for Efficient LLM Mobile Speculative Inference with Architecture-Dataflow Co-Optimization Develops a hardware-dataflow co-optimization framework using LPDDR PIM and draft token pruning for mobile LLM speculative inference. Ranked #468Peer-reviewed0/10 suggestedOlder paper (2025) Older paper (2025) — discussed this week | Small and efficient models via ieee · 12 Oct, 16:30 | On-topic Introduces a hardware-dataflow co-design using PIM and token pruning for efficient LLM mobile inference. | |
| Recursive Self-Improvement in Unified Multimodal Models The authors introduce a training loop where multimodal models use text and visual abilities alongside program execution to generate their own training data. PreprintOlder paper (2026) Older paper (2026) — discussed this week | Small and efficient models via blog · 18 Oct, 06:22 | Not relevant Focuses on recursive self-improvement training, not model compression or efficient inference. | |
| SPECGUARD:ACAMERA-MANIFOLD GAUSSIAN ATLAS ... Summary pending (next check). Could not confirmPreprint | Small and efficient models via openreview · 18 Oct, 06:18 | Waiting for check | |
| FUTURE-AWARE TRAINING-FREE KV CACHE POLICY Summary pending (next check). Could not confirmPreprint | Small and efficient models via openreview · 18 Oct, 06:18 | Waiting for check | |
| Mapping the Set-Function Geometry of Evidence Utility in ... Summary pending (next check). Could not confirmPreprint | Small and efficient models via openreview · 18 Oct, 06:18 | Waiting for check | |
| OMNI123: EXPLORING 3D NATIVE FOUNDATION MOD Summary pending (next check). Could not confirmPreprint | Small and efficient models via openreview · 18 Oct, 06:18 | Waiting for check | |
| STREAMWAM: REAL-TIME WORLD-ACTION MOD- Summary pending (next check). Could not confirmPreprint | Small and efficient models via openreview · 18 Oct, 06:18 | Waiting for check | |
| Zero-RAG: Towards Retrieval-Augmented Generation with Zero Redundant Knowledge The paper introduces a method to reduce redundant external knowledge retrieval by estimating what a language model already knows. Peer-reviewedOlder paper (2025) Older paper (2025) — discussed this week | Small and efficient models via blog · 17 Oct, 06:24 | Not relevant Focuses on RAG corpus optimization, not small models, distillation, or quantization. | |
| A Highly Efficient Design of Triple-Band Flexible Rectenna for Ambient RF Energy Harvesting in Passive IoT Applications The paper presents a flexible triple-band rectenna designed for ambient radio-frequency energy harvesting to power passive Internet of Things devices. Peer-reviewedOlder paper (2025) Older paper (2025) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant About RF energy harvesting hardware, not small or efficient AI models. | |
| Efficient Serverless Resource Allocation for MapReduce Job Based on A* Algorithm The paper introduces an improved A-star algorithm to optimize serverless resource allocation and job scheduling for MapReduce workloads. Peer-reviewedOlder paper (2024) Older paper (2024) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant Addresses cloud serverless resource allocation, not small or efficient AI models. | |
| Power and Area-Efficient Multiplier Architectures: A Comparative Study of Array, Dadda, Booth, Wallace Tree, and Vedic Multipliers The paper evaluates and compares the power consumption and resource utilization of five 64-bit multiplier architectures implemented on an FPGA platform. Peer-reviewedOlder paper (2025) Older paper (2025) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant Deals with digital hardware multipliers, not small or efficient AI models. | |
| Efficient Data Augmentation for Bioacoustics: A Noise-Enriched Bird Sound Dataset The paper proposes a noise-enrichment data augmentation methodology and curated dataset to improve bird sound recognition models under limited data conditions. Peer-reviewedOlder paper (2025) Older paper (2025) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant Focuses on bioacoustic data augmentation, not small or efficient AI models. | |
| ETPS: Efficient Two-Pass Encoding Scheme for Adaptive Live Streaming The paper presents an efficient two-pass encoding scheme using low-complexity spatial and temporal features for adaptive live video streaming. Peer-reviewedOlder paper (2022) Older paper (2022) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant Focuses on video compression codecs, not small or efficient AI models. | |
| Improved fast gauss transform and efficient kernel density estimation The paper proposes an improved fast Gauss transform algorithm for efficient kernel density estimation in computer vision applications. Peer-reviewedOlder paper (2003) Older paper (2003) — discussed this week | Small and efficient models via ieee · 17 Oct, 06:20 | Not relevant Abstract contains no technical content to judge model efficiency. |
421–480 of 2,430← PreviousNext →