Ahmed Masry Profile
Ahmed Masry

@Ahmed_Masry97

Followers
343
Following
607
Statuses
138

PhD Student @YorkUniversity | Visiting Researcher @ServiceNowRSRCH

Toronto, Ontario
Joined July 2020
Don't wanna be here? Send us removal request.
@Ahmed_Masry97
Ahmed Masry
14 hours
Doing a PhD after publishing many papers is a wild ride. One screen: launching an experiment on 64 H100s GPUs for an 8B VLM. Other screen: coding a tiny CNN for MNIST.....for a course assignment. Truly living in two eras at once πŸ˜… but it's a nice refresh πŸ˜‰
0
0
11
@Ahmed_Masry97
Ahmed Masry
2 days
RT @PerouzT: πŸš€ We just released the 3rd test split of our #RepLiQA dataset! πŸŽ‰ A unique QA dataset with reference documents that have never…
0
12
0
@Ahmed_Masry97
Ahmed Masry
3 days
I just came across two nice articles from LG AI Research and MarketTechPost featuring our work on ChartInstruct (ACL 2024) and ChartGemma (Coling 2025) πŸ“ŠπŸš€
0
2
3
@Ahmed_Masry97
Ahmed Masry
9 days
@iScienceLuvr Thanks for sharing our work! πŸ˜ƒ You can find the key highlights in this thread:
@Ahmed_Masry97
Ahmed Masry
9 days
Happy to announce AlignVLMπŸ“: a novel approach to bridging vision and language latent spaces for multimodal understanding in VLMs! πŸŒπŸ“„πŸ–ΌοΈ πŸ”— Read the paper: πŸ§΅πŸ‘‡ Thread
Tweet media one
0
0
3
@Ahmed_Masry97
Ahmed Masry
9 days
πŸ“’ Our codes and model weights will be publicly available soon! Huge thanks to our amazing collaborators! πŸ™Œ@ServiceNowRSRCH @YorkUniversity @Mila_Quebec
0
0
2
@Ahmed_Masry97
Ahmed Masry
9 days
3️⃣ We're introducing the AlignVLM models, by training Llama 3.2 1B, 3B, and 8B using our connector. We've also benchmarked it against a range of models! πŸ“Š AlignVLM outperforms all Base VLMs trained under similar configurations! πŸ“ˆ We also noticed that our models perform competitively against Instruct VLMs of similar sizes. πŸš€
Tweet media one
1
0
4
@Ahmed_Masry97
Ahmed Masry
9 days
1️⃣ Aligning visual features with language embeddings is a major bottleneck in Vision-Language Models (VLMs). Existing connectors often introduce noise, hurting performance. ❌ We propose AlignVLM, a novel method that maps vision features into a weighted average of LLM text embeddings, ensuring they remain in a space the LLM can effectively interpret. πŸŽ―βœ…
1
0
4
@Ahmed_Masry97
Ahmed Masry
10 days
@CSProfKGD unless you have another paper submission coming in a few days πŸ˜‚
1
0
1
@Ahmed_Masry97
Ahmed Masry
13 days
@CSProfKGD Yes, I know that latex comments are accessible to the public on arXiv. But manually cleaning the overleaf project was always a nightmare to me. I used to spend hours doing that.πŸ˜… I will try this tool this time instead! πŸ˜€
0
0
2
@Ahmed_Masry97
Ahmed Masry
13 days
Conference Travel βœ…οΈ ICML Submission βœ…οΈ Workshop proposal βœ…οΈ Time to focus on the ACL submission 🫑 and the projects proposals for the courses πŸ€•
@Ahmed_Masry97
Ahmed Masry
19 days
PhD life is no joke 🀯. Juggling an ICML submission, ACL submission, workshop proposal, traveling for a conference (jetlag included ✈️😴), and worrying about two courses and some TA duties πŸ“š. Adrenaline is at an all-time high πŸ˜΅β€πŸ’«
0
0
27
@Ahmed_Masry97
Ahmed Masry
15 days
@ShinDehghani Indeed, we get to travel and meet a lot of people! Also, research is very fulfulling and satisfying!
0
0
1
@Ahmed_Masry97
Ahmed Masry
19 days
@dhadfieldmenell Thanks! I will definitely take some time off to avoid getting burnt out.
0
0
1
@Ahmed_Masry97
Ahmed Masry
19 days
@GalarnykMichael Indeed, finishing the classes reduces the workload a lot, but I am sure we will have other things to stress out about :)
1
0
2
@Ahmed_Masry97
Ahmed Masry
19 days
@RajeswarSai Thanks so much for the support, Sai! I am sure the journey will be worth it in the end πŸ˜„
0
0
2