AS Intern

About Reality Defender

Reality Defender is a groundbreaking security platform offering comprehensive deepfake detection. A Y Combinator graduate, Comcast NBCUniversal LIFT Labs alumni, and backed by DCVC, Reality Defender's proactive deepfake and AI-generated content detection technology is developed by a leadership team with over 20 years of experience in applied research at the intersection of machine learning, data science, and cybersecurity.

With models defending against present and future fabrication techniques, Reality Defender is the best way to detect and deter fraudulent text, audio, and visual content, partnering with government agencies and enterprise clients to enhance security and detect fraud.

Responsibilities

• Investigate saliency and activation mapping techniques on different forms of audio representations, e.g, spectrograms and acoustic features (e.g. pitch, formants)

• Build insights for decision attribution on audio inputs, e.g., most impactful segments within an audio sample towards a classification outcome

• Prepare an explainable-AI library for generative audio detection models

• Propose novel end-to-end solutions for attribution/explainability and use them to improve detection models

• Collaborate with scientists and engineers across the organization

About You

• Currently enrolled in a Masters or PhD program in deep learning, computer vision, speech/audio processing, or a related field - Implemented and/or published peer-reviewed papers in reputable AI research venues such as CVPR, ICLR, Interspeech

• Have 1+ years of hands-on experience with model interpretation tools, e.g., saliency maps, attention maps, Grad-CAM, SHAP.

• Have 1+ years of programming experience in Python and model building in PyTorch; experience with audio models, e.g. HuBERT/wav2vec, would be a plus but not essential.

• Team player with a positive attitude and good communication skills