Mansi Phute

Title: Mansi Phute: Innovator in Vision-Language Models

Introduction

Mansi Phute is a talented inventor based in Atlanta, GA (US). She has made significant contributions to the field of vision-language models, showcasing her innovative spirit and technical expertise. Her work focuses on steering model outputs to achieve specific behavioral changes through advanced techniques.

Latest Patents

Mansi holds a patent for her invention titled "Model output steering using universal images." This patent describes a method where the output of a vision-language model (VLM) can be directed by receiving an input that includes both an image and text. The process involves compositing the image with a universal image designed to elicit specific model activations associated with a targeted behavioral change. The modified image, along with the text, is then supplied to the VLM, resulting in an output that reflects the desired behavioral change. This technique allows for steering the VLM's output without altering its internal states. The generated output can then be utilized in various applications, processes, or systems.

Career Highlights

Mansi is currently employed at Hiddenlayer, Inc., where she continues to develop her innovative ideas and contribute to the advancement of technology in her field. Her work at Hiddenlayer emphasizes the importance of integrating vision and language to create more effective models.

Collaborations

Due to space constraints, the collaborations section will be omitted.

Conclusion

Mansi Phute is a pioneering inventor whose work in vision-language models is paving the way for new advancements in technology. Her innovative patent demonstrates her ability to blend creativity with technical knowledge, making her a valuable asset in the field.