I would assume you've got Java developers as the audience. In general it is hard to provide feedback without understanding of the audience.
It looks like you've missed "Immediate Dominators" feature. I would say it is very useful, and it is not that very well known.
You do show OQL, however you miss things like "group by", "analytic functions", "joins", etc etc.
In general it is great when the presentation ends with a summary / call for action.
In other words, you probably want the listeners to apply something, or you want them to alter the way they solve problems. That is something you put to the closing slide (which was missing :( )
It would be great if you could repeat the questions. Some of the questions were hard to hear.
I would recommend to split slides and reveal the information one-by-one.
That applies for both "text-like" slides and "the ones with images".
For instance, "getting heap dump" slide has 7 bullet points, and it is kind of hard to figure out which you are describing at the moment.
It would work much better if you just used 7 slides for that and put relevant pictures/samples/etc.
In my experience, 300 clicks (e.g. small animations) per hour is a good rate for a technical talk.
The same goes for the description of "retained set" / "dominator tree".
You do use "mouse pointer" to explain "retained set", and it is hard to spot since the pointer is small, and slide background has the same color.
The audience is familiar with Java, so it makes sense to show samples of Java code, then you can show how that works in MAT.
For instance, you do show "Person" in MAT, however you do not show its code.
Trivial samples and small steps would make things easier to understand and it would make things easier to explain at the same time.
The font size was small. Well. It is OK to watch on YouTube, however I guess it was really hard to see on the presentation.
One of the tricks (works for 15" screen) is you go to fullscreen mode, and try to read slides from 1m distance.
MAT screens are hard to see as well. It would be great if MAT had a presentation view.
Then you don't really want to put important information at the bottom of the slides (as it would likely be hidden by the heads).
For instance, at 9:02 you say something really important on "a-b-c-d-e", however I guess we can never understand that because the "bottom diagram" is completely hidden by yourself.
Note: humans rarely name things like "a-b-c-d", so it would work better if you could invent a domain model for the talk (e.g. Person, Group, etc, etc)
Then you can use those items in the samples. That would make things simpler to understand.
