arXiv subjects

Arjun S. Nair

Publications and source records attributed to Arjun S. Nair.

5 recordsLinked to original sources

arXiv2026

ReLANCE: A Resource-Efficient Low-Latency Cortical Neural Acceleration Engine

We present a Cortical Neural Pool (CNP) architecture featuring a high-speed, resource-efficient CORDIC based Hodgkin-Huxley (RCHH) neuron model. Unlike shared CORDIC-based DNN approaches, the proposed neuron leverages modular and performance-optimised CORDIC stages with a latency-area trade-off. We introduce a novel Constraint-Aware Modular Parallelism (CAMP) with Precision & Stability handling to leverage maximum speedup and utilisation of hardware through hardware software co-design. The FPGA implementation of the RCHH neuron shows 24.5% LUT reduction and 35.2% improved speed, compared to SoTA designs, with 70% better normalised root mean square error (NRMSE). Furthermore, the CNP exhibits 2.85x higher throughput (12.69 GOPS) than a functionally equivalent CORDIC-based DNN engine, with only a 0.35% accuracy drop relative to the DNN counterpart on the MNIST dataset. The overall results indicate that the design shows biologically accurate, low-resource spiking neural network implementations for resource-constrained edge AI applications. The reproducibility codes are publicly available at https://github.com/mukullokhande99/CNP RCHH, facilitating rapid integration and further development by researchers.

cs.NE↗

arXiv2026

Chronicals: A High-Performance Framework for LLM Fine-Tuning with 3.51x Speedup over Unsloth

Large language model fine-tuning is bottlenecked by memory: a 7B parameter model requires 84GB--14GB for weights, 14GB for gradients, and 56GB for FP32 optimizer states--exceeding even A100-40GB capacity. We present Chronicals, an open-source training framework achieving 3.51x speedup over Unsloth through four synergistic optimizations: (1) fused Triton kernels eliminating 75% of memory traffic via RMSNorm (7x), SwiGLU (5x), and QK-RoPE (2.3x) fusion; (2) Cut Cross-Entropy reducing logit memory from 5GB to 135MB through online softmax computation; (3) LoRA+ with theoretically-derived 16x differential learning rates between adapter matrices; and (4) Best-Fit Decreasing sequence packing recovering 60-75% of compute wasted on padding. On Qwen2.5-0.5B with A100-40GB, Chronicals achieves 41,184 tokens/second for full fine-tuning versus Unsloth's 11,736 tokens/second (3.51x). For LoRA at rank 32, we reach 11,699 tokens/second versus Unsloth MAX's 2,857 tokens/second (4.10x). Critically, we discovered that Unsloth's reported 46,000 tokens/second benchmark exhibited zero gradient norms--the model was not training. We provide complete mathematical foundations: online softmax correctness proofs, FlashAttention IO complexity bounds O(N^2 d^2 M^{-1}), LoRA+ learning rate derivations from gradient magnitude analysis, and bin-packing approximation guarantees. All implementations, benchmarks, and proofs are available at https://github.com/Ajwebdevs/Chronicals with pip installation via https://pypi.org/project/chronicals/.

cs.LG↗

arXiv2025

Calibration Meets Reality: Making Machine Learning Predictions Trustworthy

Post-hoc calibration methods are widely used to improve the reliability of probabilistic predictions from machine learning models. Despite their prevalence, a comprehensive theoretical understanding of these methods remains elusive, particularly regarding their performance across different datasets and model architectures. Input features play a crucial role in shaping model predictions and, consequently, their calibration. However, the interplay between feature quality and calibration performance has not been thoroughly investigated. In this work, we present a rigorous theoretical analysis of post-hoc calibration methods, focusing on Platt scaling and isotonic regression. We derive convergence guarantees, computational complexity bounds, and finite-sample performance metrics for these methods. Furthermore, we explore the impact of feature informativeness on calibration performance through controlled synthetic experiments. Our empirical evaluation spans a diverse set of real-world datasets and model architectures, demonstrating consistent improvements in calibration metrics across various scenarios. By examining calibration performance under varying feature conditions utilizing only informative features versus complete feature spaces including noise dimensions, we provide fundamental insights into the robustness and reliability of different calibration approaches. Our findings offer practical guidelines for selecting appropriate calibration methods based on dataset characteristics and computational constraints, bridging the gap between theoretical understanding and practical implementation in uncertainty quantification. Code and experimental data are available at: https://github.com/Ajwebdevs/calibration-analysis-experiments.

cs.LG↗

arXiv2025

Disproving the Feasibility of Learned Confidence Calibration Under Binary Supervision: An Information-Theoretic Impossibility

We prove a fundamental impossibility theorem: neural networks cannot simultaneously learn well-calibrated confidence estimates with meaningful diversity when trained using binary correct/incorrect supervision. Through rigorous mathematical analysis and comprehensive empirical evaluation spanning negative reward training, symmetric loss functions, and post-hoc calibration methods, we demonstrate this is an information-theoretic constraint, not a methodological failure. Our experiments reveal universal failure patterns: negative rewards produce extreme underconfidence (ECE greater than 0.8) while destroying confidence diversity (std less than 0.05), symmetric losses fail to escape binary signal averaging, and post-hoc methods achieve calibration (ECE less than 0.02) only by compressing the confidence distribution. We formalize this as an underspecified mapping problem where binary signals cannot distinguish between different confidence levels for correct predictions: a 60 percent confident correct answer receives identical supervision to a 90 percent confident one. Crucially, our real-world validation shows 100 percent failure rate for all training methods across MNIST, Fashion-MNIST, and CIFAR-10, while post-hoc calibration's 33 percent success rate paradoxically confirms our theorem by achieving calibration through transformation rather than learning. This impossibility directly explains neural network hallucinations and establishes why post-hoc calibration is mathematically necessary, not merely convenient. We propose novel supervision paradigms using ensemble disagreement and adaptive multi-agent learning that could overcome these fundamental limitations without requiring human confidence annotations.

cs.LG↗

arXiv2025

Axi-Higgs portal Dark Matter via Wess-Zumino mechanism

We study the axion portal between the visible and the dark sector, where the dark matter is charged under an abelian extension of the Standard Model. In general, such models are anomalous and are rendered gauge invariant by a St{ü}ckelberg axion through Wess-Zumino/Green-Schwarz mechanism. Scenarios such as this naturally exist in TeV scale string theory completions of Standard Model. This axion mixes with other Goldstone bosons in the model to give a physical axi-Higgs which becomes massive upon breaking the anomalous gauge group. Such axi-Higgs fields charged under the anomalous symmetry act as mediators for the dark matter annihilation to Standard Model particles and can lead to an efficient freeze-out mechanism. Here, we show that the St{ü}ckelberg axion, and the resultant axi-Higgs, with its appropriate shift symmetry cancels the quantum anomalies and also generates the observed relic density for the dark matter. Moreover, we show that the relevant parameter space in our model, where photon production dominates, is safe from {\it Fermi}LAT, Cherenkov Telescope Array, and H.E.S.S. indirect detection experiments.

hep-ph↗