Mohamed El-Geish

Santa Clara, California, United States Contact Info
16K followers 500+ connections

Join to view profile

  • Contact Mohamed for services

    Custom Software Development and Business Consulting

About

Bio: https://elgeish.com

I'm passionate about machine learning, big data…

Articles by Mohamed

See all articles

Contributions

Activity

Join now to see all activity

Experience & Education

  • Monta AI

View Mohamed’s full experience

See their title, tenure and more.

or

By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.

Licenses & Certifications

Volunteer Experience

  • CodePath.org Graphic

    Volunteer

    CodePath.org

    - Present 4 years 1 month

    Economic Empowerment

    CodePath provides college students with no-cost coding courses, mentorship, and career support to prepare them for careers in tech. Volunteers help students with resume reviews, mock technical interviews, and giving feedback.

  • Upwardly Global Graphic

    Volunteer

    Upwardly Global

    - Present 4 years 9 months

    Economic Empowerment

    Upwardly Global supports immigrants and refugees who want to contribute their vital skills to the professional US workforce. Volunteers help students with resume reviews, mock technical interviews, and giving feedback.

  • ZAWAYA Graphic

    Member of the Board of Directors

    ZAWAYA

    - 3 years 1 month

    Arts and Culture

    ZAWAYA's mission is to preserve, produce, and promote Arabic arts in the San Francisco Bay Area. Our vision is to achieve cross-cultural communication and understanding through the universal language of the arts.

  • TechWadi.org Graphic

    Mentor

    TechWadi.org

    - 1 year 1 month

    Economic Empowerment

    Mentoring entrepreneurs who are creating their own new businesses in the tech field.

Publications

  • Gestalt: a Stacking Ensemble for SQuAD2.0

    We propose a deep-learning system -- for the SQuAD2.0 task -- that finds, or indicates the lack of, a correct answer to a question in a context paragraph. Our goal is to learn an ensemble of heterogeneous SQuAD2.0 models that, when blended properly, outperforms the best model in the ensemble per se. We created a stacking ensemble that combines top-N predictions from two models, based on ALBERT and RoBERTa, into a multiclass classification task to pick the best answer out of their predictions…

    We propose a deep-learning system -- for the SQuAD2.0 task -- that finds, or indicates the lack of, a correct answer to a question in a context paragraph. Our goal is to learn an ensemble of heterogeneous SQuAD2.0 models that, when blended properly, outperforms the best model in the ensemble per se. We created a stacking ensemble that combines top-N predictions from two models, based on ALBERT and RoBERTa, into a multiclass classification task to pick the best answer out of their predictions. We explored various ensemble configurations, input representations, and model architectures. For evaluation, we examined test-set EM and F1 scores; our best-performing ensemble incorporated a CNN-based meta-model and scored 87.117 and 90.306, respectively -- a relative improvement of 0.55% for EM and 0.61% for F1 scores, compared to the baseline performance of the best model in the ensemble, an ALBERT-based model, at 86.644 for EM and 89.760 for F1.

    See publication

  • Learning Joint Acoustic-Phonetic Word Embeddings

    Most speech recognition tasks pertain to mapping words across two modalities: acoustic and orthographic. In this work, we suggest learning encoders that map variable-length, acoustic or phonetic, sequences that represent words into fixed-dimensional vectors in a shared latent space; such that the distance between two word vectors represents how closely the two words sound. Instead of directly learning the distances between word vectors, we employ weak supervision and model a binary…

    Most speech recognition tasks pertain to mapping words across two modalities: acoustic and orthographic. In this work, we suggest learning encoders that map variable-length, acoustic or phonetic, sequences that represent words into fixed-dimensional vectors in a shared latent space; such that the distance between two word vectors represents how closely the two words sound. Instead of directly learning the distances between word vectors, we employ weak supervision and model a binary classification task to predict whether two inputs, one of each modality, represent the same word given a distance threshold. We explore various deep-learning models, bimodal contrastive losses, and techniques for mining hard negative examples such as the semi-supervised technique of self-labeling. Our best model achieves an F1 score of 0.95 for the binary classification task.

    See publication

  • Computing with Data

    Springer

    Computing with Data introduces basic computing skills designed for industry professionals without a strong computer science background. Written in an easily accessible manner, it serves as a self-study guide to survey data science and data engineering for those who aspire to start a computing career or expand on their current roles, in areas such as applied statistics, big data, machine learning, data mining, and informatics. The authors draw from their combined experience working at software…

    Computing with Data introduces basic computing skills designed for industry professionals without a strong computer science background. Written in an easily accessible manner, it serves as a self-study guide to survey data science and data engineering for those who aspire to start a computing career or expand on their current roles, in areas such as applied statistics, big data, machine learning, data mining, and informatics. The authors draw from their combined experience working at software and social network companies, on big data products at several major online retailers, as well as their experience building big data systems for an AI startup. Spanning from the basic inner workings of a computer to advanced data manipulation techniques, this book opens doors for readers to quickly explore and enhance their computing knowledge.

    Other authors
    See publication

  • Multiple-Instance, Cascaded Classification for Keyword Spotting in Narrow-Band Audio

    NIPS

    We propose using cascaded classifiers for a keyword spotting (KWS) task on narrow-band (NB), 8kHz audio acquired in non-IID environments --- a more challenging task than most state-of-the-art KWS systems face. We present a model that incorporates Deep Neural Networks (DNNs), cascading, multiple-feature representations, and multiple-instance learning. The cascaded classifiers handle the task's class imbalance and reduce power consumption on computationally-constrained devices via early…

    We propose using cascaded classifiers for a keyword spotting (KWS) task on narrow-band (NB), 8kHz audio acquired in non-IID environments --- a more challenging task than most state-of-the-art KWS systems face. We present a model that incorporates Deep Neural Networks (DNNs), cascading, multiple-feature representations, and multiple-instance learning. The cascaded classifiers handle the task's class imbalance and reduce power consumption on computationally-constrained devices via early termination. The KWS system achieves a false negative rate of 6% at an hourly false positive rate of 0.75

    Other authors
    See publication

Patents

  • Characterizing accuracy of ensemble models for automatic speech recognition by determining a predetermined number of multiple ASR engines based on their historical performance

    Issued US11380315B2

    One embodiment of the present invention sets forth a technique for analyzing a transcription of a recording. The technique includes generating features representing transcriptions produced by multiple automatic speech recognition (ASR) engines from voice activity in the recording and a best transcription of the recording produced by an ensemble model from the transcriptions. The technique also includes applying a machine learning model to the features to produce a score representing an accuracy…

    One embodiment of the present invention sets forth a technique for analyzing a transcription of a recording. The technique includes generating features representing transcriptions produced by multiple automatic speech recognition (ASR) engines from voice activity in the recording and a best transcription of the recording produced by an ensemble model from the transcriptions. The technique also includes applying a machine learning model to the features to produce a score representing an accuracy of the best transcription. The technique further includes storing the score in association with the best transcription.

    See patent

  • Ensemble modeling of automatic speech recognition output

    Issued US US11094326B2

    One embodiment of the present invention sets forth a technique for performing ensemble modeling of ASR output. The technique includes generating input to a machine learning model from snippets of voice activity in the recording and transcriptions produced by multiple automatic speech recognition (ASR) engines from the recording. The technique also includes applying the machine learning model to the input to select, based on transcriptions of the snippet produced by at least one contributor ASR…

    One embodiment of the present invention sets forth a technique for performing ensemble modeling of ASR output. The technique includes generating input to a machine learning model from snippets of voice activity in the recording and transcriptions produced by multiple automatic speech recognition (ASR) engines from the recording. The technique also includes applying the machine learning model to the input to select, based on transcriptions of the snippet produced by at least one contributor ASR engine of the multiple ASR engines and at least one selector ASR engine of the multiple ASR engines, a best transcription of the snippet from possible transcriptions of the snippet produced by the multiple ASR engines. The technique further includes storing the best transcription in association with the snippet.

    See patent

  • Characterizing accuracy of ensemble models for automatic speech recognition

    Issued US US11024315B2

    One embodiment of the present invention sets forth a technique for analyzing transcriptions of a recording. The technique includes storing per-character differences between a first set of characters from a first transcription of the recording and a second set of characters from a second transcription of the recording in a matrix with a fixed width. The technique also includes encoding the per-character differences in the matrix into a vector of the fixed width. The technique further includes…

    One embodiment of the present invention sets forth a technique for analyzing transcriptions of a recording. The technique includes storing per-character differences between a first set of characters from a first transcription of the recording and a second set of characters from a second transcription of the recording in a matrix with a fixed width. The technique also includes encoding the per-character differences in the matrix into a vector of the fixed width. The technique further includes outputting the vector as a representation of a pairwise error rate between the first transcription and the second transcription.

    See patent

  • Organizing and aggregating meetings into threaded representations

    Issued US US10972299B2

    One embodiment of the present invention sets forth a technique for organizing meeting content. The technique includes generating, from a set of available meetings, a thread comprising a collection of related meetings that share one or more attributes. The technique also includes aggregating data for the related meetings, where the data comprises metadata for the related meetings and terms included in recordings of the related meetings. The technique further includes outputting at least a…

    One embodiment of the present invention sets forth a technique for organizing meeting content. The technique includes generating, from a set of available meetings, a thread comprising a collection of related meetings that share one or more attributes. The technique also includes aggregating data for the related meetings, where the data comprises metadata for the related meetings and terms included in recordings of the related meetings. The technique further includes outputting at least a portion of the aggregated data within a summary of the thread.

    Other inventors
    See patent

  • Generating summaries and insights from meeting recordings

    Issued US US10942953B2

    One embodiment of the present invention sets forth a technique for generating a summary of a recording. The technique includes generating an index associated with the recording, wherein the index identifies a set of terms included in the recording and, for each term in the set of terms, a corresponding location of the term in the recording. The technique also includes determining categories of predefined terms to be identified in the index and identifying a first subset of the terms in the…

    One embodiment of the present invention sets forth a technique for generating a summary of a recording. The technique includes generating an index associated with the recording, wherein the index identifies a set of terms included in the recording and, for each term in the set of terms, a corresponding location of the term in the recording. The technique also includes determining categories of predefined terms to be identified in the index and identifying a first subset of the terms in the index that match a first portion of the predefined terms in the categories. The technique further includes outputting a summary of the recording comprising the locations of the first subset of terms in the recording and listings of the first subset of terms under one or more corresponding categories.

    See patent

  • Calculating efficient messaging parameters

    Issued US US10931620B2

    Methods and systems for creating optimal time limited messages is disclosed. The server system receives a limited time message request from a first member, wherein the message request identifies a second member as the intended recipient and includes an offer. The server system accesses a member profile associated with the second member. The server system determines a message expiration time recommendation based, at least in part, on member profile data of the second member. The server system…

    Methods and systems for creating optimal time limited messages is disclosed. The server system receives a limited time message request from a first member, wherein the message request identifies a second member as the intended recipient and includes an offer. The server system accesses a member profile associated with the second member. The server system determines a message expiration time recommendation based, at least in part, on member profile data of the second member. The server system receives an expiration time selection from the first member. The server system creates a limited time message based on the received message request. The server system transmits the limited time message to the second member.

    Other inventors
    See patent

  • Integrating an interactive virtual assistant into a meeting environment

    Issued US US10897368B2

    Various embodiments of the present disclosure set forth a computer-implemented method for integrating an interactive virtual assistant in a meeting environment. The method comprises receiving an invitation to participate in the meeting environment, where the invitation includes text information, and applying at least a subset of a plurality of parsers to the text information in a given order to generate a text match for each of a plurality of fields of meeting information. The method further…

    Various embodiments of the present disclosure set forth a computer-implemented method for integrating an interactive virtual assistant in a meeting environment. The method comprises receiving an invitation to participate in the meeting environment, where the invitation includes text information, and applying at least a subset of a plurality of parsers to the text information in a given order to generate a text match for each of a plurality of fields of meeting information. The method further comprises, in response, when the text matches for at least a subset of the plurality of fields of meeting information meet one or more predetermined threshold criteria, causing the interactive virtual assistant to join the meeting environment based on the text matches for the at least the subset of the plurality of fields of meeting information.

    See patent

  • Generating summaries and insights from meeting recordings

    Issued US US10860797B2

    One embodiment of the present invention sets forth a technique for generating a summary of a recording. The technique includes generating an index associated with the recording, wherein the index identifies a set of terms included in the recording and, for each term in the set of terms, a corresponding location of the term in the recording. The technique also includes determining categories of predefined terms to be identified in the index and identifying a first subset of the terms in the…

    One embodiment of the present invention sets forth a technique for generating a summary of a recording. The technique includes generating an index associated with the recording, wherein the index identifies a set of terms included in the recording and, for each term in the set of terms, a corresponding location of the term in the recording. The technique also includes determining categories of predefined terms to be identified in the index and identifying a first subset of the terms in the index that match a first portion of the predefined terms in the categories. The technique further includes outputting a summary of the recording comprising the locations of the first subset of terms in the recording and listings of the first subset of terms under one or more corresponding categories.

    See patent

  • Rerouting to an intermediate landing page

    Issued US US10726093B2

    A system and method for intermediate landing page rerouting are provided. In example embodiments, determine whether a webpage associated with a hyperlink has corresponding social network activities. Extract content from the webpage determined to have corresponding social network activities. In response to a selection of the hyperlink, reroute a web browser to an intermediate landing page. Cause presentation, at a user interface, of the extracted content and the corresponding social network…

    A system and method for intermediate landing page rerouting are provided. In example embodiments, determine whether a webpage associated with a hyperlink has corresponding social network activities. Extract content from the webpage determined to have corresponding social network activities. In response to a selection of the hyperlink, reroute a web browser to an intermediate landing page. Cause presentation, at a user interface, of the extracted content and the corresponding social network activities.

    Other inventors
    See patent

  • Chat-based interaction with an in-meeting virtual assistant

    Filed US US20210021558A1

    Chat-based interaction with an in-meeting virtual assistant may be provided. First, audio input associated with a meeting may be received. Next, an intent from the audio input may be detected. Text content associated with the audio input may then be generated in response to detecting the intent from the audio input. The text content may be displayed in a chat interface.

    Other inventors
    See patent

  • Restoring temporal coherence of previously seen ranked content

    Issued US US10261952B1

    The disclosed embodiments provide a system for restoring temporal coherence of ranked content that was previously shown to a user. During operation, the system receives a set of content items to display to the user. For each content item, the system obtains a ranking score; determines if the content item was impressed upon the user during a prior session; and, if so, adds the content item to an impressed subset of content items. The system creates a display order for the set of content items…

    The disclosed embodiments provide a system for restoring temporal coherence of ranked content that was previously shown to a user. During operation, the system receives a set of content items to display to the user. For each content item, the system obtains a ranking score; determines if the content item was impressed upon the user during a prior session; and, if so, adds the content item to an impressed subset of content items. The system creates a display order for the set of content items by: removing the impressed content items from the set of content items; ordering the set of content items according to their ranking scores; ordering the impressed subset of content items according to their display order in the prior session; and inserting the impressed subset into the set of content items at a determined insertion point. The system then displays the ordered set of content items.

    Other inventors
    See patent

  • Maximizing the propensity of content consumption according to content complexity and content consumption time

    Issued US US10062101B2

    The disclosed embodiments provide a system for maximizing the propensity of content consumption according to content complexity and content consumption time. During operation, the system receives a request from a user to access a content feed. Based on a time of receipt, the system identifies a current seasonality that corresponds to reading habits of the user during a time period encompassing the time of receipt. The system then accesses a profile for the user to obtain a reading speed of the…

    The disclosed embodiments provide a system for maximizing the propensity of content consumption according to content complexity and content consumption time. During operation, the system receives a request from a user to access a content feed. Based on a time of receipt, the system identifies a current seasonality that corresponds to reading habits of the user during a time period encompassing the time of receipt. The system then accesses a profile for the user to obtain a reading speed of the user and the reading habits of the user for the current seasonality. Next, the system: identifies a subset of content items; estimates a reading session length; and determines a complexity, with regard to the user, for each content item. The system then creates the content feed by ordering the subset of content items according to the reading session length and the complexity of each content item.

    Other inventors
    See patent

  • Obtaining and utilizing feedback for agent-assist systems

    Filed US20220383153A1

    Techniques for agent-assist systems to provide context-aware, subdocument-granularity recommended answers to agents that are attempting to answer queries of users. The agent-assist system may obtain collections of documents that include information for responding to queries, and analyze those documents to identify subdocuments that are associated with different semantics or meanings. Subsequently, any queries received can be analyzed to identify their semantics, and relevant subdocuments can be…

    Techniques for agent-assist systems to provide context-aware, subdocument-granularity recommended answers to agents that are attempting to answer queries of users. The agent-assist system may obtain collections of documents that include information for responding to queries, and analyze those documents to identify subdocuments that are associated with different semantics or meanings. Subsequently, any queries received can be analyzed to identify their semantics, and relevant subdocuments can be identified as having similar semantics. When the agent-assist system presents the agent with the relevant documents, it may highlight or otherwise indicate the relevant subdocument within the document for quick identification by the agent. Further, the agent-assist system may collect feedback from the agent and/or user to determine a relevancy of the recommended answers. The agent-assist system can use the feedback to improve the quality of the recommended answers provided to the agents.

    Other inventors
    See patent

  • Providing agent-assist, context-aware recommendations

    Filed US20220156298A1

    Techniques for agent-assist systems to provide context-aware, subdocument-granularity recommended answers to agents that are attempting to answer queries of users. The agent-assist system may obtain collections of documents that include information for responding to queries, and analyze those documents to identify subdocuments that are associated with different semantics or meanings. Subsequently, any queries received can be analyzed to identify their semantics, and relevant subdocuments can be…

    Techniques for agent-assist systems to provide context-aware, subdocument-granularity recommended answers to agents that are attempting to answer queries of users. The agent-assist system may obtain collections of documents that include information for responding to queries, and analyze those documents to identify subdocuments that are associated with different semantics or meanings. Subsequently, any queries received can be analyzed to identify their semantics, and relevant subdocuments can be identified as having similar semantics. When the agent-assist system presents the agent with the relevant documents, it may highlight or otherwise indicate the relevant subdocument within the document for quick identification by the agent. Further, the agent-assist system may collect feedback from the agent and/or user to determine a relevancy of the recommended answers. The agent-assist system can use the feedback to improve the quality of the recommended answers provided to the agents.

    Other inventors
    See patent

Courses

  • Artificial Intelligence: Principles and Technique

    CS221

  • Deep Learning

    CS230

  • Machine Learning

    CS229

  • Natural Language Processing with Deep Learning

    CS224n

Projects

Honors & Awards

  • Recognized as a High Potential employee

    Microsoft

    Microsoft Leadership Bench (9/2012 - 11/2014)

Test Scores

  • Thanaweya Amma (University-Entrance Examination)

    Score: 100.1%

    I placed first nationwide in the 2002 Thanaweya Amma (university-entrance examination) for science and math concentration (and third overall out of roughly half a million students).

Languages

  • Arabic

    Native or bilingual proficiency

  • English

    Native or bilingual proficiency

Recommendations received

11 people have recommended Mohamed

Join now to view

More activity by Mohamed

View Mohamed’s full profile

  • See who you know in common
  • Get introduced
  • Contact Mohamed directly
Join to view full profile

Explore collaborative articles

We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.

Explore More

Others named Mohamed El-Geish

1 other named Mohamed El-Geish is on LinkedIn

See others named Mohamed El-Geish

Add new skills with these courses

See all courses