Searcharxiv⌕ Search

arXiv subjects

Nikhil Krishna

Publications and source records attributed to Nikhil Krishna.

5 recordsLinked to original sources

WM-Quant: Rollout-Aware Post-Training Quantization for JEPA-Style World Models

We present WM-Quant, to our knowledge, the first post-training quantization (PTQ) framework designed specifically for JEPA-style world models. Although world models typically feature relatively few parameters, action planning algorithms often require thousands of inferences, which can quickly become computationally expensive. PTQ is a natural way to reduce this cost. However, we show that in world models using autoregressive model predictive control (MPC) rollouts, quantization error can compound with each predictor step, resulting in action rankings from the quantized model becoming near-random. WM-Quant addresses this with three components that build on existing PTQ methods: early-encoder precision, control-path preservation, and rollout-aware scaling. Early-encoder precision protects accurate latent state and goal information to mitigate long-term error propagation. Control-path preservation allows the world model predictor to distinguish between fine-grained actions, significantly improving action rankings and execution. Rollout-aware scaling jointly calibrates encoder and per-step predictor activation ranges with respect to an entire rollout rather than a single prediction, reducing the cumulative rollout error. Across multiple world models, environments, and PTQ methods, WM-Quant substantially improves planning success at low bit-widths, approaching full-precision performance at W4A4.

cs.LG↗

Sensitivity of the photon-induced processes to the proton charge radius

We study the exclusive production of dileptons in proton--proton collisions as a probe of the proton charge-radius scale. Using a dipole form factor model, we compare the conventional choice of $Λ^2=0.71$~GeV$^2$ with PDG test scenarios corresponding to $r_p=0.8751 \pm 0.0061$~fm and $r_p=0.84087 \pm 0.00039$~fm, representative of the historical large-radius from CODATA and smaller-radius from muonic-hydrogen determinations, respectively. The sensitivity is greatest at large dilepton invariant masses and forward/backward rapidity. Fitting to the current ATLAS and CMS data within the adopted model gives $Λ^2 = 0.465 \pm 0.056~\mathrm{GeV}^2$, corresponding to an effective radius $r_p = 1.002 \pm 0.038~\mathrm{fm}$, which indicates non-trivial sensitivity on the proton radius scale, but is not yet a definitive solution to the proton radius puzzle. While this result depends on the theoretical model used in the calculation, it shows that exclusive dilepton production at LHC can provide complementary model-dependent information on the proton electromagnetic forms factor.

hep-ph↗

Node-wise Feature Encoding for Neural Performance Prediction

As neural networks are increasingly deployed on resource constrained edge devices, accurate prediction of latency and energy is critical for efficient neural architecture search. Existing GNN and transformer based predictors achieve strong results but largely ignore node-level computational cost, limiting their ability to model performance critical operations. To address this, we introduce FeatureFormer, a neural performance predictor that incorporates explicit node-wise encodings of FLOPs, parameter counts, and memory proxies within a gated graph attention architecture. We also present NNEQ, a new large-scale energy consumption dataset that enables unified evaluation of latency and energy prediction. Extensive experiments demonstrate that FeatureFormer achieves state-of-the-art performance across both metrics, including challenging out-of-domain settings. Finally, we show that the proposed encoding is broadly applicable and consistently improves existing predictors with negligible overhead.

cs.LG↗

From photons to dikaon -- theoretical insights into $K^+K^-$ production in nuclear collisions

The production of charged kaon pairs in ultraperipheral heavy-ion collisions can proceed via photoproduction (gamma-Pomeron interaction) or via photon-photon fusion. An important contribution to this process arises from the decays of scalar, tensor and vector mesons. This study provides a consistent description of K+K- production at both the elementary level (gamma gamma -> meson -> K+K-) and the nuclear level (Pb Pb -> Pb Pb K+K-). The gamma gamma fusion cross section is compared with experimental results from Belle, TPC/Two-Gamma and ARGUS. A comparison with existing midrapidity measurements is presented, together with theoretical predictions for ultraperipheral Pb-Pb collisions at sqrt(sNN) = 5.02 TeV.

hep-ph↗

How Do Expectations Affect Learning About Fundamentals? Some Experimental Evidence

We test how individuals with incorrect beliefs about their ability learn about an external parameter (`fundamental') when they cannot separately identify the effects of their ability, actions, and the parameter on their output. Heidhues et al. (2018) argue that learning makes overconfident individuals worse off as their beliefs about the fundamental get less accurate, causing them to take worse actions. In our experiment, subjects take incorrectly-marked tests, and we measure how they learn about the marker's accuracy over time. Overconfident subjects put in less effort, and their beliefs about the marker's accuracy got worse, as they learnt. Beliefs about the proportion of correct answers marked as correct fell by 0.05 over the experiment. We find no effect in underconfident subjects.

econ.GN↗