arXiv subjects

Jake Lee

Publications and source records attributed to Jake Lee.

At least 19 recordsLinked to original sources

arXiv2026

Triton for MTIA: Bridging the Programming Model Gaps for Custom AI Accelerators

The rapid growth in machine learning workloads has fueled the proliferation of custom accelerator architectures. Designed from the ground up, these accelerators often expose programming models that are distinct from GPUs. While hyperscalers and AI chip startups continue to innovate in this space, achieving broad operator coverage to support diverse models remains a major challenge. Additionally, an easy-to-use, high-level kernel programming language is important for rapid iteration of models and kernels. Triton, together with TorchInductor, addresses these issues on GPUs, but its viability on accelerators with different programming models has yet to be established. In this work, we present the first production-scale application of Triton on a custom ML accelerator, MTIA-2i, developed by Meta. To support MTIA-2i, we develop a new compiler backend that targets it, introduce enhancements to TorchInductor code generation, and propose minimal language extensions that expose MTIA-specific architectural features. We demonstrate that Triton-MTIA kernels achieve performance competitive with expert-tuned C++ implementations. Leveraging these development efficiency gains, we successfully deployed manually written and Inductor-generated Triton kernels in production across approximately 60 different model types, accounting for 50% of layers and 47% of non-GEMM execution time for these models. Our results provide compelling evidence that DSLs like Triton can bridge the programming model gaps between ML frameworks, kernels, and custom accelerators, enabling rapid innovation and efficient deployment at scale.

cs.PL↗

arXiv2026

Fully Automatic Trace Gas Plume Detection

Future imaging spectrometers will expand contemporary data volumes by orders of magnitude, requiring automated methods to upscale labor-intensive detection of trace gas point sources. Here we present a fully-automated approach that achieves operational performance for plume detection and labelling without human participation. Our method combines machine learning (ML)-based morphological analysis with physics-based spectroscopic model fitting. We deploy it on data from the EMIT imaging spectrometer, operating in two modes. First, we present a "daily digest" that runs automatically on all downlinked data, flagging the largest events for immediate response. The daily digest demonstrates that a significant fraction of the largest plumes can be detected automatically with negligible false positives. This represents a significant new high-water mark in plume detection accuracy. Second, we use it for retrospective analysis to find plumes that were missed by the existing human review process. We observe that at least 25% of large plumes may have been passed over in the existing workflow due to confirmation bias and ambiguity in the visual cues used by human reviewers. Finally, we extend detection to three understudied trace gases: NH3, NO2 and the first observations of carbon monoxide (CO) plume in EMIT imagery.

cs.LG↗

arXiv2026

Adaptive MSD-Splitting: Enhancing C4.5 and Random Forests for Skewed Continuous Attributes

The discretization of continuous numerical attributes remains a persistent computational bottleneck in the induction of decision trees, particularly as dataset dimensions scale. Building upon the recently proposed MSD-Splitting technique -- which bins continuous data using the empirical mean and standard deviation to dramatically improve the efficiency and accuracy of the C4.5 algorithm -- we introduce Adaptive MSD-Splitting (AMSD). While standard MSD-Splitting is highly effective for approximately symmetric distributions, its rigid adherence to fixed one-standard-deviation cutoffs can lead to catastrophic information loss in highly skewed data, a common artifact in real-world biomedical and financial datasets. AMSD addresses this by dynamically adjusting the standard deviation multiplier based on feature skewness, narrowing intervals in dense regions to preserve discriminative resolution. Furthermore, we integrate AMSD into ensemble methods, specifically presenting the Random Forest-AMSD (RF-AMSD) framework. Empirical evaluations on the Census Income, Heart Disease, Breast Cancer, and Forest Covertype datasets demonstrate that AMSD yields a 2-4% accuracy improvement over standard MSD-Splitting, while maintaining near-identical O(N) time complexity reductions compared to the O(N log N) exhaustive search. Our Random Forest extension achieves state-of-the-art accuracy at a fraction of standard computational costs, confirming the viability of adaptive statistical binning in large-scale ensemble learning architectures.

cs.LG↗

arXiv2025

Markerless Augmented Reality Registration for Surgical Guidance: A Multi-Anatomy Clinical Accuracy Study

Purpose: In this paper, we develop and clinically evaluate a depth-only, markerless augmented reality (AR) registration pipeline on a head-mounted display, and assess accuracy across small or low-curvature anatomies in real-life operative settings. Methods: On HoloLens 2, we align Articulated HAnd Tracking (AHAT) depth to Computed Tomography (CT)-derived skin meshes via (i) depth-bias correction, (ii) brief human-in-the-loop initialization, (iii) global and local registration. We validated the surface-tracing error metric by comparing "skin-to-bone" relative distances to CT ground truth on leg and foot models, using an AR-tracked tool. We then performed seven intraoperative target trials (feet x2, ear x3, leg x2) during the initial stage of fibula free-flap harvest and mandibular reconstruction surgery, and collected 500+ data per trial. Results: Preclinical validation showed tight agreement between AR-traced and CT distances (leg: median |Delta d| 0.78 mm, RMSE 0.97 mm; feet: 0.80 mm, 1.20 mm). Clinically, per-point error had a median of 3.9 mm. Median errors by anatomy were 3.2 mm (feet), 4.3 mm (ear), and 5.3 mm (lower leg), with 5 mm coverage 92-95%, 84-90%, and 72-86%, respectively. Feet vs. lower leg differed significantly (Delta median ~1.1 mm; p < 0.001). Conclusion: A depth-only, markerless AR pipeline on HMDs achieved ~3-4 mm median error across feet, ear, and lower leg in live surgical settings without fiducials, approaching typical clinical error thresholds for moderate-risk tasks. Human-guided initialization plus global-to-local registration enabled accurate alignment on small or low-curvature targets, improving the clinical readiness of markerless AR guidance.

cs.CV↗

arXiv2025

Agentic generative AI for media content discovery at the national football league

Generative AI has unlocked new possibilities in content discovery and management. Through collaboration with the National Football League (NFL), we demonstrate how a generative-AI based workflow enables media researchers and analysts to query relevant historical plays using natural language rather than traditional filter-and-click interfaces. The agentic workflow takes a user query as input, breaks it into elements, and translates them into the underlying database query language. Accuracy and latency are further improved through carefully designed semantic caching. The solution achieves over 95 percent accuracy and reduces the average time to find relevant videos from 10 minutes to 30 seconds, significantly increasing the NFL's operational efficiency and allowing users to focus on producing creative content and engaging storylines.

cs.AI↗

arXiv2023

Onboard Science Instrument Autonomy for the Detection of Microscopy Biosignatures on the Ocean Worlds Life Surveyor

The quest to find extraterrestrial life is a critical scientific endeavor with civilization-level implications. Icy moons in our solar system are promising targets for exploration because their liquid oceans make them potential habitats for microscopic life. However, the lack of a precise definition of life poses a fundamental challenge to formulating detection strategies. To increase the chances of unambiguous detection, a suite of complementary instruments must sample multiple independent biosignatures (e.g., composition, motility/behavior, and visible structure). Such an instrument suite could generate 10,000x more raw data than is possible to transmit from distant ocean worlds like Enceladus or Europa. To address this bandwidth limitation, Onboard Science Instrument Autonomy (OSIA) is an emerging discipline of flight systems capable of evaluating, summarizing, and prioritizing observational instrument data to maximize science return. We describe two OSIA implementations developed as part of the Ocean Worlds Life Surveyor (OWLS) prototype instrument suite at the Jet Propulsion Laboratory. The first identifies life-like motion in digital holographic microscopy videos, and the second identifies cellular structure and composition via innate and dye-induced fluorescence. Flight-like requirements and computational constraints were used to lower barriers to infusion, similar to those available on the Mars helicopter, "Ingenuity." We evaluated the OSIA's performance using simulated and laboratory data and conducted a live field test at the hypersaline Mono Lake planetary analog site. Our study demonstrates the potential of OSIA for enabling biosignature detection and provides insights and lessons learned for future mission concepts aimed at exploring the outer solar system.

astro-ph.IM↗

arXiv2021

Mars Image Content Classification: Three Years of NASA Deployment and Recent Advances

The NASA Planetary Data System hosts millions of images acquired from the planet Mars. To help users quickly find images of interest, we have developed and deployed content-based classification and search capabilities for Mars orbital and surface images. The deployed systems are publicly accessible using the PDS Image Atlas. We describe the process of training, evaluating, calibrating, and deploying updates to two CNN classifiers for images collected by Mars missions. We also report on three years of deployment including usage statistics, lessons learned, and plans for the future.

cs.LG↗

arXiv2020

What Does CNN Shift Invariance Look Like? A Visualization Study

Feature extraction with convolutional neural networks (CNNs) is a popular method to represent images for machine learning tasks. These representations seek to capture global image content, and ideally should be independent of geometric transformations. We focus on measuring and visualizing the shift invariance of extracted features from popular off-the-shelf CNN models. We present the results of three experiments comparing representations of millions of images with exhaustively shifted objects, examining both local invariance (within a few pixels) and global invariance (across the image frame). We conclude that features extracted from popular networks are not globally invariant, and that biases and artifacts exist within this variance. Additionally, we determine that anti-aliased models significantly improve local invariance but do not impact global invariance. Finally, we provide a code repository for experiment reproduction, as well as a website to interact with our results at https://jakehlee.github.io/visualize-invariance.

cs.LG↗

arXiv2018

Interpretable Discovery in Large Image Data Sets

Automated detection of new, interesting, unusual, or anomalous images within large data sets has great value for applications from surveillance (e.g., airport security) to science (observations that don't fit a given theory can lead to new discoveries). Many image data analysis systems are turning to convolutional neural networks (CNNs) to represent image content due to their success in achieving high classification accuracy rates. However, CNN representations are notoriously difficult for humans to interpret. We describe a new strategy that combines novelty detection with CNN image features to achieve rapid discovery with interpretable explanations of novel image content. We applied this technique to familiar images from ImageNet as well as to a scientific image collection from planetary science.

cs.LG↗

arXiv2007

Leptogenesis in a seesaw model with Fritzsch type lepton mass matrices

We investigate how the baryon asymmetry of our universe via leptogenesis can be achieved within the framework of the seesaw model with Fritzsch type lepton mass matrices proposed by Fukugita $et. al$. We study the cases with CP-violating phases in charged lepton Yukawa matrix, however, with and without Dirac neutrino Yukawa phases. We consider both flavor independent and flavor dependent leptogenesis, and demonstrate how they lead to different amounts of lepton asymmetries in detail. In particular, it is shown that flavor dependent leptogenesis in this model can be worked out only when the CP phases in Dirac neutrino Yukawa matrix become zero at the GUT scale. In addition to the CP phases, for successful leptogenesis in the model it is required that the degeneracy of the heavy Majorana neutrino mass spectrum should be broken and we also show that the breakdown of the degeneracy can be radiatively induced.

hep-ph↗

arXiv2007

Correlation between lepton flavor violation and B_{(d,s)} - \bar{B}_{(d,s)} mixing in SUSY GUT

Motivated by the recent measurements of the B_s-\bar{B}_s mass difference from the D0 and CDF collaborations, we probe new physics effects in the B_q-\bar{B}_q mixing within the context of the supersymmetric grand unified model (SUSY GUT). We find that new physics effects in B_{s(d)}-\bar{B}_{s(d)} mixing lead to the correlated information in the branching fractions of the lepton flavor violating decays, which may serve as a test of the SUSY GUT. We also discuss the implication of such new physics effects on the quark-lepton complementarity in the neutrino mixings.

hep-ph↗

arXiv2007

$μ-τ$ Symmetry and Radiatively Generated Leptogenesis

We consider a $μ-τ$ symmetry in neutrino sectors realized at GUT scale in the context of a seesaw model. In our scenario, the exact $μ-τ$ symmetry realized in the basis where the charged lepton and heavy Majorana neutrino mass matrices are diagonal leads to vanishing lepton asymmetries. We find that, in the minimal supersymmetric extension of the seesaw model with large $\tanβ$, the renormalization group (RG) evolution from GUT scale to seesaw scale can induce a successful leptogenesis even without introducing any symmetry breaking terms by hand, whereas such RG effects lead to tiny deviations of $θ_{23}$ and $θ_{13}$ from $π/4$ and zero, respectively. It is shown that the right amount of the baryon asymmetry $η_B$ can be achieved via so-called resonant leptogenesis, which can be realized at rather low seesaw scale with large $\tanβ$ in our scenario so that the well-known gravitino problem is safely avoided.

hep-ph↗

arXiv2006

The intrinsic and oscillated astrophysical neutrino flavor ratios

The pp interactions taking place in the cosmos around us are a source of the astrophysical neutrinos of all the three flavors. In these interactions, the electron and the muon neutrinos mainly come from the production and the decay of the π^{\pm} mesons, whereas the tau neutrinos mainly come from the production and the decay of the D^{\pm}_{S} mesons. We estimate the three intrinsic neutrino flavor ratios for 1 GeV < E < 10^{12} GeV in the pp interactions and found them to be 1 : 2 : 3\times 10^{-5}. We study the effects of neutrino oscillations on these intrinsic ratios. We point out that the three ratios become 1 : 1 : 1 if L(pc)/E(GeV)> 10^{-10} in the presence of neutrino oscillations, where L is the distance to the astrophysical neutrino source in units of parsecs.

hep-ph↗

arXiv2006

Systematic Investigation of Possibilities for New Physics Effects in b --> s Penguin Processes

Although recent experimental results in b-->s penguin process seem to be roughly consistent with the standard model predictions, there may be still large possibilities of new physics hiding in this processes. Therefore, here we investigate systematically the potential new physics effects that may appear in time-dependent CP asymmetries of B --> phi K^0, B--> eta^\prime K^0 and B--> K^0 π^0 decay modes, by classifying the cases for the values of the mixing-induced indirect CP asymmetries, S_{phi K^0}, S_{eta^\prime K^0}, S_{K^0 pi^0} which are compared to S_{J/psi K^0}. We also show that several B_s decay modes may help to resolve the ambiguities in such an analysis. Through combining analysis with the time-dependent CP asymmetries of B_s decay modes such as B_s --> phi eta^\prime, B_s--> eta^\prime pi^0 and B_s --> K^0 bar{K}^0, we can determine where the new CP phases precisely come from.

hep-ph↗

arXiv2006

Exclusive B-> M ν\barν (M= π, K, ρ, K^*) Decays and Leptophobic Z^\prime Model

We consider the exclusive flavor changing neutral current processes B -> M ν\barν (M= π, K, ρ, K^*) in the leptophobic Z^\prime model, in which the charged leptons do not couple to the extra Z^\prime boson. We find that these exclusive modes are very effective to constrain the leptophobic Z^\prime model. In the leptophobic Z^\prime model, additional right-handed neutrinos are introduced and they can contribute to the missing energy signal in B -> M + E_missing decays. Through the explicit calculations, we obtain quite stringent bounds on the model parameters, |U_{sb}^{Z^\prime}| \leq 0.29 and |U_{db}^{Z^\prime}| \leq 0.61, from the already existing experimental data. We also briefly discuss an interesting subject of massive right-handed neutrinos, which might be connected with the dark matter problem.

hep-ph↗

arXiv2006

Phased Breaking of mu-tau symmetry and Leptogenesis

Non-vanishing U_{e3} has been theoretically related to a certain flavor symmetry breaking in the neutrino sector. We propose a scenario to break the mu-tau symmetry so as to accommodate the non-vanishing U_{e3}. Our scenario is constructed in the context of a seesaw model, and the mu-tau symmetry breaking is achieved by introducing a CP phase in the Dirac Yukawa matrix. We also show how the deviation of theta_{23} from the maximal mixing and non-vanishing U_{e3} depend on the CP phase. Neutrino mixings and the neutrino mass-squared differences are discussed, and the amplitude in neutrinoless double beta decay m_{ee} are also predicted. We found that a tiny breaking of the mu-tau symmetry due to mass splitting between two degenerate heavy Majorana neutrinos on top of the Dirac CP phase can lead to successful leptogenesis. We examine how leptogenesis can be related with low energy neutrino measurement, and show that our predictions for U_{e3} and m_{ee} can be constrained by the current observation of baryon asymmetry.

hep-ph↗

arXiv2005

Quark-Lepton Complementarity with Renormalization Effects through Threshold Corrections

The recent experimental measurements of the solar neutrino mixing angle $θ_{sol}$ and the Cabibbo mixing angle $θ_C$ reveal a surprising relation, $ θ_{sol}+θ_C \simeq \fracπ{4} $. We review that while this empirical relation has been interpreted as a support of the idea of grand unification, it may be merely accidental in the sense that reproducing the relation at a low energy in the framework of grand unification may depend strongly on the renormalization effects whose size can vary with the choice of parameter space. We note that the lepton mixing matrix derived from quark-lepton unification can lead to a shift of the complementarity relation at low energy. While the renormalization group effects generally lead to additive contribution on top of the shift, we show that the threshold corrections which may exist in some intermediate scale new physics such as supersymmetric standard model can diminish it, so we can achieve the complementarity relation at a low energy.

hep-ph↗

arXiv2005

Importance of Threshold Corrections in Quark-Lepton Complementarity

The recent experimental measurements of the solar neutrino mixing angle $θ_{sol}$ and the Cabibbo mixing angle $θ_C$ reveal a surprising relation, $ θ_{sol}+θ_C \simeq \fracπ{4}$. We note that the lepton mixing matrix derived from quark-lepton unification can lead to a shift of the complementarity relation at low energy. While the renormalization group effects generally lead to additive contribution on top of the shift, in this letter, we show that the threshold corrections which may exist in some intermediate scale new physics such as supersymmetric standard model can diminish it, so we can achieve the complementarity relation at a low energy. Finally, we discuss a possibility to achieve the complementarity relation at a high energy by taking particular form of non-symmetric form of down Yukawa matrix.

hep-ph↗