Valency
Join the waitlist

Research dossierORCID 0000-0003-2988-2361

Khalid M. Mosalam

University of California, Berkeley · 2 affiliations on record

What this is

Valency read your published work and assembled this dossier automatically. Every figure links to the real paper behind it.

257 works, 8,196 citations, h-index 46, and your recent machine-learning preprints are already cited in structural-health-monitoring reviews.

Most of your career sits in earthquake and structural engineering journals, so the slice we hold is the newer one: the work where you point machine learning at the safety of the built environment. The damage-assessment GAN, the fiber-optic sensing of a reinforced-concrete connection, the differentiable graph-network agent that hunts for a building's worst fire-initiation point, the language model that writes disaster reconnaissance reports, and the social-physics model of how hazards spread through populations. Each takes a structural-safety question that is expensive to answer by simulation or by hand and builds a method that answers it faster without giving up on rigor. Here is how that part of your output reads from outside.

257works in corpus
2020–2026active
Generated June 11, 2026
02
Your research program

Read as one thread, not a list

We hold only the recent corner of your output, the part where machine learning meets structural safety, so read this as a window on that corner and not the whole record. The damage-assessment GAN is the most cited at 93, and it keeps getting picked up by structural-health-monitoring reviews and by other groups fighting the same low-data, class-imbalance problem. The fiber-optic sensing work runs on a separate track, closer to the lab and the instrumented specimen, and it travels among people who measure real concrete. The newer pieces, the fire-sensitive-point agent, the reconnaissance language model, the social-amplification model, are too recent to have citations yet, though each carries the work into a new specialty the way the GAN did.

Where your output lives
  • Machine Learningcs.LG33%
  • Computer Visioncs.CV17%
  • Image and Video Processingeess.IV17%
  • Physics and Societyphysics.soc-ph17%
  • Computation and Languagecs.CL17%
Career, by the numbers
257works
8,196citations
46h-index

Those 8,196 citations are the full weight of a long earthquake-engineering career, most of it in journals we do not index. Inside the recent machine-learning work we do hold, the damage-assessment GAN carries the load at 93 citations, with the fiber-optic sensing study adding a smaller, steadier stream.

The papers carrying the load
03
Where your work travels

Across fields and sources

Your damage-assessment GAN travels furthest. It lands in the big structural-health-monitoring surveys, the sensor-technology review at 364 citations and the deep-learning-for-SHM review at 80, and it occasionally jumps domains entirely, as when a self-supervised EEG paper cites it for learning from scarce labels. The newest citers stay close to home: a 2026 study brings damage identification to concrete arch dams, and an aerospace-SHM review folds your work into its account of data scarcity. The fiber-optic sensing line keeps its own company, cited by people comparing core-versus-surface strain sensors and by the DeJong group's later distributed-sensing work on reinforced concrete.

Highest-impact citations
  1. A Systematic Review of Advanced Sensor Technologies for Non-Destructive Testing and Structural Health Monitoring

    2023Sensors (Basel, Switzerland)364 citesHassani, Sahar, Dackermann, Ulrike

    A widely cited survey of sensing for non-destructive testing and structural health monitoring that places your damage-assessment GAN in the data-analytics half of the field.

    indexed
  2. Self-Supervised Learning for Electroencephalography

    2024IEEE Transactions on Neural Networks and Learning Systems282 citesRafiei, Mohammad H, Gauthier, Lynne V, et al.

    Reaches across domains to cite your low-data semi-supervised approach as prior art for learning from scarce labels, here on brain signals rather than cracked concrete.

    indexed
  3. Deep Learning for Structural Health Monitoring: Data, Algorithms, Applications, Challenges, and Trends

    2023Sensors (Basel, Switzerland)80 citesJia, Jing, Li, Ying

    A systematic review of deep learning for structural health monitoring that catalogs your GAN among the generative methods for the image and vibration data that dominate the field.

    indexed
Most recent citations
  1. Dynamic Response-Based Safety Monitoring and Damage Identification of Concrete Arch Dams via PSO-LSTM

    2026Sensors (Basel, Switzerland)0 citesQiu, Jianchun, He, Wenqin, et al.

    Carries the data-driven damage-identification idea onto concrete arch dams, using an LSTM tuned by particle-swarm optimization on seismic dynamic responses to flag structural damage.

    indexed
  2. Machine Learning for Structural Health Monitoring of Aerospace Structures: A Review

    2025Sensors (Basel, Switzerland)3 citesScarselli, Gennaro, Nicassio, Francesco

    Pulls your damage-assessment work into an aerospace-SHM review, treating data scarcity and interpretability as the shared open problems across civil and aerospace monitoring.

    indexed
Citing fields beyond your own
17%Physics and Societyphysics.soc-ph
e.g.Social Amplification Dominates Collective Hazard Response

When a paper cites you but we don't hold its full text, we flag it, so you can tell first-hand sources from second-hand mentions.

04
Your collaboration network

Two centers of gravity

Inside the work we hold, your collaborators sort into three tight knots. Guanren Zhou anchors the social-physics group, where you publish with Broccardo, Sornette, Ziqi Wang, and Chu on how hazards ripple through populations, and that group co-authors densely with one another. Yuqing Gao is your steadiest partner on the vision side, on the damage GAN and then the reconnaissance language model, and Gao's own output is accelerating into computer vision and AI. Pengyuan Zhai started with you on the extreme-events text-analytics paper alongside Alicia Tsai and Laurent El Ghaoui, then pulled hardest away: Zhai's record now runs into white-box deep learning with Yi Ma and Shengbang Tong and a move to Harvard, well outside structural engineering. One caution on this section. Your hundreds of long-standing structural-engineering co-authors do not appear here because that work lives in journals we do not index, so this is the network of your recent machine-learning papers, not your full career.

Khalid M. MosalamYuqing Gao · 2Guanren Zhou · 2Pengyuan Zhai · 2
Clusters
  • Social-physics hazard-response group
  • Vision-based damage and reconnaissance
  • Extreme-events text analytics
Most frequent collaborators
  1. Yuqing Gaocs.CV2papers
  2. Guanren Zhouphysics.soc-ph2papers
  3. Pengyuan Zhaics.LG2papers
  4. Marco Broccardo1papers
  5. Didier Sornette1papers
  6. Ziqi Wang1papers
  7. Xiaolei Chu1papers
  8. Selim Gunay1papers
  9. Laurent El Ghaoui1papers
  10. Yuan Xinjie1papers
  11. Alicia Y. Tsai1papers
05
People you should meet

Close to your work, not yet in your network

Three researchers keep landing next to your recent directions, none at your institutions and none with a paper co-authored with you. Mohsen Zaker Esteghamati works the trust-and-deployment question your methods quietly depend on. Yazhou Xie scales seismic-response surrogates from one structure to whole portfolios. Yuecheng Cai builds the graph-network stand-in for finite-element analysis that your fire-sensitive-point work also relies on. Any of them would bring a surrogate-modeling or trustworthy-machine-learning angle that reinforces yours from outside your current circle.

Mohsen Zaker Esteghamati

Established
explainable surrogate models for performance-based seismic designthe gap between developing and deploying machine learning in structural engineering

He builds explainable surrogate models for seismic design and writes directly about why machine learning so often fails to reach deployment in structural engineering, which is the trust question sitting underneath your own fire-sensitive-point and damage-assessment work.

He works at Utah State and Virginia Tech, shares no institution with you, and has no paper co-authored with you.

Yazhou Xie

Established
deep-learning surrogates for the seismic response of bridge portfolios

His surrogate models predict probabilistic seismic responses across whole portfolios of bridges from partial information, a scaling-up of exactly the simulation-replacement idea your differentiable thermal-agent work runs at the single-building level.

He is at McGill, with earlier stints at Rice and UCLA, shares no institution with you, and has no paper co-authored with you.

Yuecheng Cai

Rising
graph neural network surrogates for finite-element structural analysis

He represents stiffened structural panels as heterogeneous graphs so a network can stand in for finite-element analysis under varied loads and boundary conditions, the same graph-as-surrogate move your most-fire-sensitive-point agent makes, from a different starting point.

He is at the University of British Columbia, shares no institution with you, and has no paper co-authored with you.

06
Frontiers in your fields, last 90 days

What just landed next to your work

Across the last 90 days your fields are converging on the surrogate. Structural and earthquake engineers are replacing expensive finite-element runs with graph networks and transfer-learned models, vision-language agents are writing the disaster reports you started automating, and reviewers are now asking what these surrogates cost in energy and whether their early errors propagate. The social-physics side keeps probing when collective response tips, and a fresh review finally pairs fiber-optic sensing with machine learning, stitching together the two lines you have been working from opposite ends.

07
Where your next paper should go

Five directions, grounded in your work

All five sit between what you have already built and where these fields are now pushing, toward methods an engineer can actually trust to make a safety call. The worst-case-surrogate and uncertainty-calibration ideas carry your fire-safety and deployment instincts into seismic engineering, and the second points straight at Mohsen Zaker Esteghamati. The generalizable-GAN and grounded-reconnaissance ideas harden your two most-developed lines against the failure each is most exposed to. The last reconnects your social-physics work to the structural risk models that should be feeding it.

Every source below is real and linked. The directions, and the reading of them, are generated. Check anything you would lean on against your own knowledge of the field.

  1. 01

    From worst fire to worst earthquake: differentiable surrogates for the most critical loading scenario

    Your most-fire-sensitive-point work trains a differentiable graph network as a stand-in for finite-element fire analysis so you can search for the single worst fire-initiation point instead of simulating all of them. Seismic engineers are building graph and transfer-learned surrogates for response prediction, but they use them for forward prediction or optimization, not to hunt for a structure's worst-case excitation.

    The gap

    A search for a differentiable graph-network surrogate used to identify a worst-case seismic scenario returned only your own most-fire-sensitive-point paper doing that worst-case search. The neighboring work (a graph structural simulator, a graph-network-plus-genetic-algorithm optimizer) does forward response or parameter tuning, none of it the worst-case-excitation hunt.

    First step

    Recast the most-fire-sensitive-point method for seismic loading: train a differentiable graph-network agent on nonlinear time-history responses and use its gradients to find the ground-motion or input that maximizes interstory drift, the earthquake analog of the worst fire.

    Bridgescs.LGcs.CE
  2. 02

    Damage-assessment GANs that generalize across structures, not just datasets

    Your BSS-GAN solved low-data, class-imbalanced damage classification with balanced semi-supervised training. The frontier keeps re-solving the same imbalance one dataset at a time, with new attention networks and reweighted GANs, each tuned to its own benchmark. Nobody is asking the augmentation to transfer the way a real inspection program needs, from one structure type or site to the next.

    The gap

    Recent damage-detection papers (MS-SSE-Net, the class-imbalance augmentation and reweighted-ACGAN follow-ons that cite your GAN) each fight imbalance within a single dataset. None pairs the generative augmentation with domain adaptation so the trained detector holds up on a structure or imaging condition it never saw.

    First step

    Extend BSS-GAN with a domain-adaptation objective and test it cross-structure: train on one inventory of cracked-and-spalled concrete and measure recall on a different structure type without retraining.

    Bridgescs.LGeess.IV
  3. 03

    Reconnaissance reports a language model can be held to: grounding and verification

    Your disaster-reconnaissance language model turns field images and metadata into written reports. The wider field is racing to the same place with multi-agent vision-language systems, and the most visible of them openly warns that an early misread propagates into the final report. Your pipeline already has the trained damage classifiers that could check the language model's claims, but nothing closes that loop.

    The gap

    DisasTeller, the leading large-vision-language post-disaster framework, explicitly names error propagation and the need for human validation as its open problem. None of the current reconnaissance systems grounds the language model's narrative against the underlying damage-classifier outputs to catch and flag inconsistent claims.

    First step

    Add a verification layer to your reconnaissance pipeline that cross-checks each sentence of the generated report against the damage-state, material, and severity outputs of the vision models, and surfaces any claim the evidence does not support.

    Bridgescs.CVcs.CL
  4. 04

    Coupling the physical hazard to the social response: structural risk meets emotional amplification

    Your social-amplification model shows collective emotion outrunning the physical threat in most U.S. states during COVID-19, using epidemiological severity as the hazard input. For earthquakes and fires you already build calibrated physical-risk models. Joining the two would let the amplification model run on a real structural-damage forcing rather than a generic hazard proxy.

    The gap

    A search for work coupling physical hazard severity with social-media emotional amplification returned your own paper on top, with the rest being social-media flood monitoring, emotion analysis, and resilience surveys. None feeds a calibrated structural or engineering damage model in as the physical forcing term of an emotional-contagion model.

    First step

    Replace the generic hazard input in the social-amplification model with output from a structural damage or loss model for one well-instrumented earthquake, and test whether grounding the physical term sharpens the prediction of where collective response amplifies.

    Bridgesphysics.soc-phcs.CE
  5. 05

    Uncertainty-calibrated surrogates for performance-based earthquake engineering

    Surrogate models for seismic response are multiplying, and they hand downstream risk assessments point predictions treated as ground truth. The same community is starting to say out loud that the harder problem is trusting these models enough to deploy them. Your worst-case and residual-style diagnostics are a natural way to tell when a surrogate is operating outside what it can be trusted to predict, but no one has paired calibrated uncertainty with that kind of diagnostic for performance-based design.

    The gap

    The frontier shows seismic-response surrogates proliferating (a transfer-learning surrogate for a 20-story frame, explainable surrogates for performance-based design), while a survey of deploying machine learning in structural engineering names trust and deployment as the open gap. None of them equips a performance-based earthquake-engineering surrogate with calibrated uncertainty plus an out-of-distribution diagnostic.

    First step

    Build an uncertainty-calibrated surrogate for nonlinear time-history response with Mohsen Zaker Esteghamati, and add an out-of-distribution flag so the model reports honest confidence bounds and refuses ground motions it should not extrapolate to.

    Bridgescs.LGcs.CE
See it live

Run this live, Khalid, on any researcher in any field.

What you just read is a snapshot. Inside Valency you walk the citation graph yourself, re-run the network, and chase these directions or wherever your own questions lead, in real time or async through your agents and your co-scientist. Join the waitlist and we’ll send you an invite.

Join the waitlist