Eric Delmelle · YouTube · 6:02
This 150-second whiteboard lecture introduces multiple-objective optimization in spatial applications, first defining problems with more than one objective and giving examples such as minimizing cost, maximizing access, maximizing coverage, and minimizing travel. The lecturer emphasizes that objectives are often but not always conflicting, then narrows to a bi-objective case for easier visualization. The central method is a weighted-sum scalarization, J = ( · OBJ_1) + () · OBJ_2, implemented by tabulating different values of and mapping the resulting solutions into the OBJ_1-OBJ_2 plane. The clip ends by showing that varying the weight produces a trade-off curve rather than a single universal optimum. This 150-second whiteboard segment introduces multi-objective optimization through a two-objective minimization plot. The instructor explains that the objectives conflict, defines the Pareto front as the curve linking non-dominated solutions, distinguishes dominated and unattainable points, shows the weighted-sum scalarization J=(·OBJ_1)+()·OBJ_2, and uses a school cost-versus-travel example to motivate gradual improvement of feasible solutions toward the Pareto front. This 62-second whiteboard segment explains how heuristic methods are used in multi-objective optimization when searching near the Pareto front. The lecturer names simulated annealing, tabu search, and genetic algorithms, then argues that even if a heuristic does not land exactly on the Pareto front, producing many feasible solutions very close to it is still valuable. While saying this, he draws a cluster of small circles just inside the dashed frontier on the right-hand objective-space plot. He further explains that one such near-optimal solution can be presented to a stakeholder to illustrate the structure of a good solution, even if it is not perfectly optimal. The clip closes by saying these ideas will be revisited later in the course when heuristic techniques and other optimization methods are covered.
Use the learning inspector for key ideas and moments, or open the reading tabs for the complete notes.
Generated from the video's visuals and explanation; not verbatim speech.
The clip opens on a title card naming the topic as multiple objective optimization and the Pareto front, preparing a whiteboard explanation rather than a worked numerical problem.
The lecturer begins by stating that the subject is multiple-objective optimization and situates it within spatial optimization. The board heading “Multiple - objective optimization” and the line “more than one objective function” define the setting: unlike earlier single-objective examples, the decision maker now tracks several goals at once.
He then lists concrete objective types—minimize cost, maximize access, maximize coverage, minimize travel—and uses them to motivate why several criteria matter in facility and service-location problems. The accompanying note “generally conflictive but not always” is important: the lecture does not claim universal opposition, only that conflict is common.
To make the geometry tractable, the lecturer reduces the discussion to two objectives. This is not presented as a loss of generality in theory, but as a visualization choice: with three or more objectives the picture becomes harder to draw and interpret.
He names the two criteria OBJ_1 and OBJ_2 and frames the task as finding an optimal combination. The school example makes the tension explicit: fewer schools can reduce one cost-like objective, while more schools can reduce student travel. This example supports the later need for a systematic way to compare compromises.
The central formula is then introduced: J = ( · OBJ_1) + () · OBJ_2. In words, the first objective receives weight and the second receives the complementary weight . The lecture treats J as a scalarized objective that summarizes a pair of competing criteria into one number for a given weight choice.
A brief transition slide, “Optimizing different functions,” separates the conceptual setup from the operational demonstration on the board.
Returning to the blackboard, the lecturer points to a table with columns , J, OBJ_1, and OBJ_2. The visible weight values are .99, .9, .75, .5, .25, and .1. He explains the procedure: choose a weight, compute J from the weighted sum, and record the solution’s objective values. The table is therefore the computational scaffold for exploring the family of weighted problems.
He then connects the table to the upper-left graph. For each weight, the corresponding solution is mapped into the OBJ_1-OBJ_2 plane. As these points are collected, they trace the purple curve the lecturer calls a trade-off curve. The visual message is that changing the weight does not merely change a number; it moves the optimizer to a different compromise between the two objectives.
By the end of the clip, the lecture has established the core conceptual chain: multiple objectives create potential conflict; a weighted sum converts the pair into a single scalar criterion; varying the weight generates a set of solutions; and plotting those solutions reveals the trade-off structure in objective space.
The segment opens on a blackboard graph with horizontal axis OBJ_1 and vertical axis OBJ_2. The instructor states the core issue: if he minimizes one objective as much as possible, it happens “to the detriment” of the other. This is the defining feature of a conflicting multi-objective problem.
A title card announces “The pareto front,” signaling a shift from the general trade-off statement to the geometric object that organizes the best compromises.
Returning to the board, he points to the purple curve and defines it verbally as the Pareto front. He says it links all the non-dominated solutions. On the drawing, the white points on that curve are labeled “non-dominated Solutions,” so the concept is introduced visually before any formal dominance inequality is written.
He then isolates a red point below and to the left of the frontier and calls it an “unattainable solution.” His wording is explicit: “We cannot find that solution. It’s simply impossible.” This establishes that not every attractive point in the objective-value plane corresponds to an achievable decision.
Next he turns to the remaining plotted points, shown in yellow and labeled “dominated solutions.” He characterizes them informally as “not very good.” In context, these are attainable points that fail to lie on the Pareto front, so better trade-off alternatives exist.
A second title card appears: “Getting from an acceptable solution to something closer to the pareto front.” This frames the rest of the clip as a methodological discussion rather than just terminology.
The instructor redraws a simpler version of the objective plot on the right side of the board, again with OBJ_2 vertical and OBJ_1 horizontal. He restates that both objectives are to be minimized and that reducing one raises the other. The repeated axes and frontier-like curve reinforce that the lesson is about trade-offs in objective-value space.
To make the abstraction concrete, he gives a school-system example. If OBJ_1 is the cost of running the school system and OBJ_2 is total student travel, then the cheapest extreme case is one school serving everyone. That lowers cost but forces long trips, increasing travel. Conversely, opening many schools reduces student travel but makes cost worse. The example is a direct instantiation of the earlier conflict claim.
He then addresses practical solution behavior. First, he says it can be hard to land exactly on the Pareto front and may take a long time. Next, on the new sketch, he marks the region above and to the right of the dashed frontier as “feasible,” while indicating that the red point outside that region is not attainable. Finally, he adds several small circles in the feasible region and explains that one often begins with feasible but non-optimal solutions and gradually improves them so they get closer to the Pareto front.
The lecturer is already discussing the Pareto front in a multi-objective optimization setting. On the board, the right-hand graph is labeled with axes "obj 1" and "obj 2", with a dashed frontier curve and a region marked "feasible".
He introduces heuristic methods by name: simulated annealing, tabu search, and genetic algorithms. A text overlay simultaneously lists these methods under the heading "heuristics:" and indicates that there are many others.
The key point is that a heuristic may sometimes reach the Pareto front itself. In the lecturer's wording, that is the case in which the method can get to an optimum.
He then turns to the more flexible practical outcome: if the heuristic does not land exactly on the frontier, it may still return many feasible solutions that are very close to it. As he says this, he draws several small circles just inside the dashed curve, visually representing a cluster of near-optimal candidates.
The goal, in his explanation, is to populate as many nearby possibilities as possible. The growing set of plotted points shows that usefulness is not limited to exact frontier membership; proximity plus multiplicity is already valuable.
Next, he points to individual drawn solutions and argues that even a solution that is not perfectly optimal can still be communicated to a stakeholder. The purpose is not to claim exact optimality, but to show what the structure of a good solution would look like.
He closes by saying this was only a brief explanation of how one gets to such solutions, and that the topic will be revisited later in the semester when heuristic techniques and other optimization methods are discussed in more detail.
The video defines multiple-objective optimization as an optimization problem with more than one objective function. It contrasts this with earlier single-objective examples and situates the topic in spatial optimization, where several goals may need to be considered together.
Typical objectives listed on the board are minimizing cost, maximizing access, maximizing coverage, and minimizing travel. The lecturer states that such objectives are generally conflicting, but not always, so conflict is treated as common rather than universal.
The lecturer narrows the discussion to a bi-objective case because three, four, or five objectives become harder and harder to visualize. The restriction is pedagogical and geometric, making the trade-off structure easier to plot in a plane.
For two objectives, the lecture combines them into a single aggregate objective J by weighting OBJ_1 with and OBJ_2 with . This turns the multi-objective problem into a family of single-objective problems indexed by the chosen weight.
The lower-left table lists candidate values of and columns for J, OBJ_1, and OBJ_2. The method is to vary , compute J from the weighted sum, and record the resulting objective values for each solution.
Mapping the solutions obtained for different weights into the OBJ_1-OBJ_2 plane produces a purple curve. The lecturer calls this a trade-off curve, emphasizing that different weights lead to different compromise solutions rather than one fixed optimum.
The clip begins by defining the central difficulty of multi-objective optimization: when two objectives are both minimized, improving one typically worsens the other. The instructor calls these “conflicting objectives.”
The purple curve on the board is identified as the Pareto front. It is described as the line linking all non-dominated solutions, i.e., the best attainable trade-off points in the plotted objective space.
These are the white points lying on the Pareto front. In this introductory presentation they are defined geometrically as the solutions connected by the frontier curve.
A red point below/left of the frontier is labeled unattainable. The instructor says such a solution cannot be found and is “simply impossible,” distinguishing it from merely poor feasible solutions.
The yellow points away from the Pareto front are called dominated solutions. The instructor says they are “not very good,” meaning they are attainable but inferior to points on the frontier.
The board displays a single combined objective formed from the two original objectives using weights that sum to one. This is the standard scalarization approach shown alongside the Pareto discussion.
A concrete example sets OBJ_1 as school-system cost and OBJ_2 as total student travel. One school minimizes cost but maximizes travel; many schools reduce travel but increase cost, illustrating the trade-off directly.
On the second sketch, the region above/right of the dashed curve is marked feasible. The instructor explains that practical methods often first produce feasible but non-optimal solutions, then improve them gradually until they lie close to the Pareto front.
The clip identifies heuristic methods that can be used to search for good solutions in multi-objective optimization. The lecturer explicitly names simulated annealing, tabu search, and genetic algorithms, and an on-screen overlay repeats these names with an ellipsis indicating many other methods exist.
According to the lecturer, a heuristic sometimes gets directly onto the Pareto front. In that case, he describes the outcome as reaching an optimum. This is presented as one possible result of heuristic search, not as a guarantee for every run.
If the heuristic does not land exactly on the Pareto front, the lecturer says it can still produce many feasible solutions that are very close to it. He treats this as an acceptable and even desirable outcome, emphasizing proximity to the frontier rather than exact membership.
The practical goal described in the clip is to generate as many nearby candidate solutions as possible. This is illustrated visually by drawing multiple small circles just inside the dashed Pareto-front curve on the objective-space plot.
The lecturer argues that a solution need not be perfectly optimal to be worth showing to a stakeholder. A near-front solution can still communicate the structure of a good solution and help decision makers understand what a reasonable candidate looks like.
The segment ends by placing this discussion in course context: the lecturer says these ideas will be revisited later in the semester when heuristic techniques and other optimization methods are covered more fully.
Explore conditions, steps and evidence. Supplementary explanations are labeled separately from content shown in the video.
The board shows OBJ_1 as the horizontal axis label in the upper-left graph and as a column header in the lower-left table.
The lecturer refers to “one objective, right? Objective one.”
OBJ_1
First objective function, plotted on the horizontal axis of the trade-off graph and listed as a column in the weight table.
objective-function value
The board shows OBJ_2 as the vertical axis label in the upper-left graph and as a column header in the lower-left table.
The lecturer says “and a second one, objective two.”
OBJ_2
Second objective function, plotted on the vertical axis of the trade-off graph and listed as a column in the weight table.
objective-function value
The boxed equation contains multiplying OBJ_1, and the table header is with values .99, .9, .75, .5, .25, .1.
The lecturer says each selected objective is multiplied by a weight representing its importance.
Weight assigned to the first objective in the weighted-sum formulation; varied across rows of the table.
weight parameter
The boxed equation contains () multiplying OBJ_2.
The lecturer says the second weight is typically one minus that weight.
The symbol is not written explicitly in the boxed formula; it is represented there as ().
Weight assigned to the second objective, stated verbally as .
weight parameter
The boxed equation begins with J = ... and the lower-left table has a column labeled J.
The lecturer says “an objective function J here is going to be a weighted sum of objective one and two.”
J
Aggregate weighted objective function used to rank or map solutions for each chosen weight.
scalarized objective value
A purple curve connects several marked points in the OBJ_1-OBJ_2 plane.
The lecturer says mapping solutions for different weights yields “this sort of trade-off curve.”
trade-off curve
Curve in the objective space formed by optimal solutions obtained as the weight changes.
set of solution points in objective space
The board repeatedly labels the horizontal axis as OBJ_1 and the instructor says “objective one.”
At 00:00:06.000-00:00:13.000 he says minimizing objective two is at the detriment of objective one.
OBJ_1
First objective value plotted on the horizontal axis; in the worked example it is cost to be minimized.
numeric objective value
The board labels the vertical axis as OBJ_2 with a nearby “min” annotation.
At 00:00:00.000-00:00:06.000 he says minimizing objective one is at the cost of objective two.
OBJ_2
Second objective value plotted on the vertical axis; in the worked example it is total travel by students, also minimized.
numeric objective value
Bottom-right boxed formula reads J = ( · OBJ_1) + () · OBJ_2.
A table below the graph has column headers , J, OBJ_1, OBJ_2.
J
Scalarized combined objective formed as a weighted sum of the two objectives.
scalar objective value
Boxed formula uses and ().
Left-side annotations list =.99, .9, .75, .5, .25, .1, .01, .001.
Weight assigned to OBJ_1 in the scalarized objective; the complementary weight on OBJ_2 is .
weight parameter shown from near 0 to near 1
At 00:02:05.000-00:02:13.000 he says “all of this here is the feasible space… All of this here is feasible.”
He writes “feasible” inside the region above/right of the dashed curve on the second graph.
feasible
Region of attainable solution points in the objective-value plane.
region in the (OBJ_1, OBJ_2) plane
The right-hand graph has a vertical axis labeled "obj 1".
obj 1
First objective-function coordinate in the right-hand objective-space plot.
Axis label in the plotted objective space; no numeric range is stated.
The lecturer introduces “multiple objective optimization” and says there may be more than one objective.
The board heading reads “Multiple - objective optimization” followed by “more than one objective function.”
The video defines multiple-objective optimization as an optimization setting with more than one objective function, contrasting it with earlier examples that had only a single objective.
There is more than one objective function to consider.
The lecturer lists objectives such as minimizing cost, maximizing access, maximizing coverage, and minimizing travel, then says they are generally conflicting but not always.
The board lists “minimize cost,” “maximize access,” “maximize coverage,” and “minimize travel,” with a brace note “generally conflictive but not always.”
The lecture gives applied examples of objectives in spatial optimization and states that multiple objectives often pull in opposite directions, although this conflict is not guaranteed in every case.
Applies to multi-objective problems where several goals are considered simultaneously.
The lecturer says the method is called a weighted objective and explains that each objective is multiplied by a weight representing its importance.
The boxed equation on the board is J = ( · OBJ_1) + () · OBJ_2.
An overlay labels this as “generic formulation (w=weight)” and another overlay marks the factors as “weights.”
For a bi-objective problem, the video combines the two objectives into a single scalar objective J by weighting OBJ_1 with and OBJ_2 with , then varying to explore different solutions.
Two objectives are being combined.
The second weight is taken as .
Different values of are tried to see how the solution changes.
The lecturer says the basis is to start with a table listing different values for , remembering that is , and then computing J as a weighted sum.
The lower-left table has headers , J, OBJ_1, OBJ_2 and visible values .99, .9, .75, .5, .25, .1.
The lecturer points between the table and the graph while explaining that each weight maps to a solution point.
The table cells under J, OBJ_1, and OBJ_2 are shown as dashes in the visible frames, so no numerical entries are actually filled in during this clip.
The video presents a practical procedure: list candidate weights in a table, compute the aggregate objective J from the weighted sum, and record the corresponding objective values OBJ_1 and OBJ_2 for each weight.
Use a set of candidate weights .
Compute J for each weight.
Record the resulting solution’s OBJ_1 and OBJ_2 values.
The lecturer says that after mapping solutions for different weights, “you end up with this sort of trade-off curve.”
A purple curve in the upper-left graph connects several marked solution points in the OBJ_1-OBJ_2 plane.
As the weight varies, the corresponding optimal solutions trace a curve in objective space; the video calls this resulting curve a trade-off curve.
At least two objectives are plotted against each other.
Solutions are generated for multiple weight values.
At 00:00:00.000-00:00:15.000 he explains that minimizing one objective harms the other and calls these conflicting objectives.
Overlay text states “minimizing objective 1 is at the cost of increasing objective 2. We want to minimize both of them.”
The clip defines the central issue of multi-objective optimization: when two objectives are both to be minimized, improving one generally worsens the other. The instructor explicitly names this tension “conflicting objectives.”
There are at least two objectives.
Both objectives are being minimized in the example.
The objectives trade off against each other.
At 00:00:23.000-00:00:31.000 he says, “Now this purple line is called the Pareto front… It’s basically linking all the non-dominated solutions.”
A purple curve connects several white points labeled “non-dominated Solutions.”
Title card at 00:00:15.000-00:00:22.000 reads “The pareto front.”
The Pareto front is presented as the curve that links all non-dominated solutions in the objective-value plot. Visually it is the lower-left boundary of the attainable set in this minimization problem.
Objectives are plotted in the (OBJ_1, OBJ_2) plane.
The example uses minimization of both objectives.
White points along the purple curve are labeled “non-dominated Solutions.”
At 00:00:23.000-00:00:31.000 he identifies the purple line as linking all the non-dominated solutions.
The clip does not state the formal dominance inequality definition.
Non-dominated solutions are the plotted points that lie on the Pareto front. In this introductory treatment they are characterized visually as the best attainable trade-off points rather than by a formal dominance relation.
They are attainable solutions.
They lie on the Pareto front in the example.
At 00:00:31.000-00:00:43.000 he says the red/yellow point is an unattainable solution and “We cannot find that solution. It’s simply impossible.”
A red point below/left of the purple curve is labeled “unattainable Solution.”
An unattainable solution is a point in the objective-value plane that cannot actually be achieved by any feasible decision. In the drawing it lies beyond the attainable frontier, below and to the left of the Pareto front.
It is represented in objective-value space.
It is outside the feasible/attainable region.
At 00:00:43.000-00:00:49.000 he says, “What about all the other ones? Those are called dominated solutions. So those are not very good.”
Yellow points away from the purple curve are labeled “dominated solutions.”
The clip does not formally define domination by comparing two solution vectors.
Dominated solutions are attainable points that are not on the Pareto front. The instructor characterizes them informally as “not very good,” meaning there exist better trade-off points available.
They are attainable but not on the Pareto front in the example.
Boxed formula on the board reads J = ( · OBJ_1) + () · OBJ_2.
A table lists values .99, .9, .75, .5, .25, .1, .01, .001 alongside columns for J, OBJ_1, OBJ_2.
The clip shows the scalarization formula but does not derive why varying traces the Pareto front.
The board presents a scalar objective J obtained by weighting the two objectives: J=(·OBJ_1)+()·OBJ_2. This is the standard weighted-sum method for turning a two-objective minimization problem into a single-objective one.
Two objectives are being combined.
Weights sum to 1 because the second weight is written as .
The example is for minimization of both objectives.
At 00:02:05.000-00:02:13.000 he says “all of this here is the feasible space… All of this here is feasible.”
On the second graph he writes “feasible” in the region above/right of the dashed curve.
The feasible space is the set of attainable points in the plotted objective-value plane. In the second sketch it is the region on or above/right of the dashed frontier curve, while points below/left are not attainable.
The plot is in objective-value coordinates.
The example concerns attainable solutions of a multi-objective problem.
The lecturer says, “Generally these objectives are conflicting, but not always.”
The board note beside the objective list reads “generally conflictive but not always.”
In multi-objective optimization, different objectives are generally in conflict, though conflict is not guaranteed in every case.
There is more than one objective function.
Generally, but not always.
The lecturer says one can go to three, four, or five objectives, “but then it’s harder and harder to visualize. So let’s just take two.”
An overlay states: “I take the example of a bi-objective weighted problem; it will be easier to visualize.”
The lecture restricts the example to two objectives because higher-dimensional objective sets become progressively harder to visualize.
The discussion is about visualizing multi-objective solutions.
For the purposes of visualization in this example.
The lecturer says that for each weight they map the solution, and then “you end up with this sort of trade-off curve.”
The upper-left graph shows a purple curve through multiple solution points labeled in terms of OBJ_1 and OBJ_2.
The clip states the result visually and verbally but does not prove that every weighted-sum optimum lies on the displayed curve.
When solutions are computed for different weights in the two-objective weighted-sum formulation, the resulting points form a trade-off curve in the OBJ_1-OBJ_2 plane.
There are two objectives.
Solutions are computed for multiple values of using J = ( · OBJ_1) + () · OBJ_2.
For the family of weights considered in the example.
At 00:00:00.000-00:00:15.000 he states that minimizing objective one is at the cost of objective two, and vice versa, producing conflicting objectives.
Overlay text repeats the same claim.
In the example, minimizing OBJ_1 increases OBJ_2, and minimizing OBJ_2 increases OBJ_1; therefore the objectives conflict.
Both OBJ_1 and OBJ_2 are to be minimized.
The plotted solutions exhibit a trade-off.
For the illustrated family of solutions, improving one objective comes at the expense of the other.
At 00:00:23.000-00:00:31.000 he says the purple line is the Pareto front and that it links all the non-dominated solutions.
The Pareto front is the curve connecting all non-dominated solutions in the objective-value plot.
The plotted points include non-dominated solutions.
The problem is shown in two objective dimensions.
All non-dominated solutions shown on the board lie on the Pareto front.
At 00:02:00.000-00:02:05.000 he says it is very hard sometimes to get solutions on the Pareto front and may take a long time.
No proof or algorithmic explanation is given in this clip.
Obtaining solutions exactly on the Pareto front may be hard and time-consuming.
One is solving the multi-objective problem in practice.
Sometimes, in practical solution processes.
The speaker says that sometimes heuristics may get to the Pareto front and that this is the case where a heuristic could get to an optimum.
The clip does not state conditions under which a heuristic is guaranteed to reach the Pareto front.
Heuristic methods such as simulated annealing, tabu search, and genetic algorithms can sometimes reach the Pareto front, and in that case they can reach an optimum.
A heuristic search method is being applied to a multi-objective optimization problem.
Some heuristic runs may succeed; the claim is not universal for all runs.
The speaker says that if heuristics do not land exactly on the Pareto front, having many very close solutions is acceptable and "that's what you want."
This is presented as the lecturer's practical criterion for usefulness, not as a formally proved theorem.
If a heuristic does not place solutions exactly on the Pareto front, producing many feasible solutions very close to that front is still a desirable outcome.
The problem is multi-objective.
The plotted points are feasible solutions in objective space.
For heuristic search outcomes discussed in this clip, closeness to the Pareto front is treated as sufficient practical value.
The speaker says a solution that is not perfectly optimal can still be presented to a stakeholder to show what the structure of the solution would look like.
The clip does not quantify how close to optimal is close enough for stakeholder use.
A solution that is not perfectly optimal can still be useful because it can be presented to stakeholders to illustrate the structure of a good solution.
The solution is near the Pareto front or otherwise meaningful in the objective space.
There is a stakeholder who needs an interpretable candidate solution.
At least some non-optimal solutions have practical communicative value.
The lecturer explains that each objective is multiplied by a weight representing importance, and the second weight is typically one minus the first.
The board displays J = ( · OBJ_1) + () · OBJ_2.
The lecturer combines the two objectives into one scalar objective by assigning weight to OBJ_1 and weight to OBJ_2.
This is the weighted-sum formulation stated directly on the board and in the audio.
The multi-objective problem is converted into a single scalar objective J whose value depends on the chosen weight .
The lecturer says to start with a table of different values, compute J, and map the solution for each weight, ending with a trade-off curve.
The lower-left table lists values and the upper-left graph shows multiple solution points connected by a purple curve.
The actual numerical mapping from each row of the table to a specific plotted point is not completed on screen in this clip.
The table provides several candidate weights for the first objective.
These values are visible in the column of the lower-left table.
For each chosen weight, the aggregate objective is computed from the two objective values.
This is the formula stated in the boxed equation and repeated verbally.
Each optimized solution is represented by its pair of objective values and plotted in the objective-space graph.
The lecturer says they map the solution for each weight and points between the table and the graph.
Collecting the plotted solutions as varies produces the displayed trade-off curve.
The lecturer explicitly calls the resulting shape a trade-off curve while indicating the purple curve.
Changing the weight generates a sequence of solutions whose objective pairs trace a trade-off curve in the OBJ_1-OBJ_2 plane.
At 00:01:32.000-00:02:00.000 he gives the school-system example linking cost and student travel.
Set OBJ_1 to be the cost of a school system and OBJ_2 to be total travel by students.
These are the two objectives named in the spoken example.
Consider the worst-case scenario for cost minimization: only one school is open.
This is the extreme case used verbally to make the trade-off clear.
With only one school, students must travel from very far away, so OBJ_2 becomes large.
Direct consequence stated by the instructor.
Conversely, if many schools are open, OBJ_1 is bad/high but OBJ_2 decreases because students travel less.
Stated in the audio at 00:01:50.000-00:02:00.000.
The example demonstrates opposite movement of the two objectives, supporting the claim that they are conflicting.
The speaker first says heuristics may get to the Pareto front, then says if not, one can still obtain many solutions very close to it.
The drawing shifts from referring to the dashed frontier itself to adding many circles just inside that frontier.
Start from a heuristic method applied to a multi-objective problem.
The lecturer introduces heuristics as search techniques for this setting.
Outcome 1: the heuristic reaches the Pareto front, which the speaker describes as reaching an optimum.
Directly stated in the audio around 00:02--00:15.
Outcome 2: the heuristic does not land exactly on the front, but yields many feasible solutions very close to it.
Directly stated in the audio around 00:18--00:28 and illustrated by drawing clustered circles near the dashed curve.
In the second outcome, the practical goal becomes populating many nearby possibilities rather than requiring exact frontier membership.
Stated in the audio around 00:28--00:32.
Heuristic search may either hit the Pareto front directly or produce a useful cluster of near-front solutions; in the latter case, closeness and multiplicity are the practical objectives.
The speaker moves from discussing near-front solutions to saying one can present such a solution to a stakeholder even if it is not perfectly optimal.
He points to specific drawn circles while explaining this use.
Take one of the drawn solutions that is near, but not exactly on, the Pareto front.
The speaker refers to "a solution like this one" while pointing at the plotted points.
Acknowledge that the solution may not be perfectly optimal.
Explicitly stated in the audio around 00:36--00:43.
Use the solution anyway to show the stakeholder what the structure of the solution would look like.
Explicitly stated in the audio around 00:43--00:50.
A near-front but non-optimal solution can still serve as a concrete illustrative candidate for decision makers.
The lecturer gives a school example: keeping the number of schools low conflicts with making students travel as little as possible.
Illustrate why two objectives can conflict in a facility-location setting.
One objective is to keep the number of schools low.
Another objective is to minimize student travel.
Explain the trade-off between fewer facilities and shorter travel.
If there is only one school, students have to travel longer distances.
Stated directly by the lecturer.
If there are many schools, students travel less.
Stated directly by the lecturer.
Fewer schools reduce one objective but worsen the other, showing a direct conflict between cost-like and travel-like goals.
The example matches the board note that multiple objectives are generally conflicting but not always.
At 00:01:32.000-00:02:00.000 the instructor says: “If for instance objective one is to minimize the cost of a school system… that is going to be at the cost of the total travel by the students…”
Illustrate how minimizing one objective can worsen another in a real-world setting.
OBJ_1 is cost of a school system.
OBJ_2 is total travel by students.
Both are to be minimized.
Show the trade-off between cost and student travel.
Take the extreme low-cost case with only one school open.
This is the verbal worst-case scenario used in the example.
Then students must travel from very far away, increasing total travel.
Stated directly in the audio.
Take the opposite case with many schools open.
Used to show the reverse direction of the trade-off.
Then cost is high but total student travel goes down.
Stated directly in the audio.
Lower cost pushes total student travel up; more schools lower student travel but raise cost.
The example matches the earlier general statement that minimizing one objective is at the detriment of the other.
The lecturer uses heuristic methods and the right-hand objective-space sketch as the running example.
He draws multiple small circles near the dashed Pareto-front curve to represent candidate solutions.
No numerical objective values or explicit problem data are provided.
Show how heuristic methods behave when searching a multi-objective problem represented in objective space with a Pareto front.
A right-hand plot with axes "obj 1" and "obj 2".
A dashed curve representing the Pareto front.
A region labeled "feasible".
Named heuristics: simulated annealing, tabu search, genetic algorithms.
Explain what kinds of solution sets heuristics may return and why near-front solutions remain useful.
Consider a heuristic such as simulated annealing, tabu search, or genetic algorithms applied to the multi-objective problem.
The speaker explicitly names these methods as examples of heuristics.
Sometimes the heuristic reaches the Pareto front itself.
Stated in the audio around 00:02--00:15.
If it does not, it can still produce many feasible solutions very close to the front, which the lecturer draws as clustered circles just inside the dashed curve.
Stated in the audio around 00:18--00:32 and matched by the drawing action.
One of these near-front solutions can be selected and shown to a stakeholder as an example of the likely solution structure, despite not being perfectly optimal.
Stated in the audio around 00:32--00:50 while the speaker points to individual plotted points.
The example concludes that heuristic search may yield either exact Pareto-front points or, more commonly in the lecturer's framing, a useful set of nearby feasible solutions that can still communicate solution structure.
The spoken explanation matches the visual sequence: the dashed frontier is referenced first, then multiple nearby circles are added, then individual circles are pointed to during the stakeholder discussion.
The blackboard is divided into an upper-left graph, a lower-left table, and right-side text with a boxed formula.
Upper-left OBJ_1-OBJ_2 graph
Lower-left weight table
Right-side heading and objective list
Boxed weighted-sum formula
The lecturer points among the graph, table, and formula as the explanation progresses.
The graph axes remain OBJ_1 horizontally and OBJ_2 vertically.
The boxed formula remains J = ( · OBJ_1) + () · OBJ_2.
The layout links the conceptual definition, the scalarizing formula, the tabular weight search, and the resulting objective-space picture.
A purple curve connects several marked points in the OBJ_1-OBJ_2 plane; labels include dominated solutions, non-dominated solutions, and unattainable solution.
The lecturer traces along the curve with the pointer while discussing mapping solutions for different weights.
The exact coordinates of the plotted points are not readable.
Horizontal axis OBJ_1
Vertical axis OBJ_2
Purple curve
Marked solution points
Labels for dominated and non-dominated regions
The pointer moves along the curve to indicate how different weights correspond to different solution locations.
The curve is drawn in the two-objective plane rather than in decision-variable space.
The visual shows that varying the weight produces a family of solutions forming a frontier-like curve rather than a single isolated optimum.
The lower-left table has headers , J, OBJ_1, OBJ_2 and visible entries .99, .9, .75, .5, .25, .1.
The lecturer points at the table while explaining that one starts with different values for .
The cells under J, OBJ_1, and OBJ_2 appear as dashes in the visible frames, so no completed numerical rows are shown.
Column
Column J
Column OBJ_1
Column OBJ_2
Rows of candidate weights
Attention shifts from the weight column to the formula and then to the graph.
The table is organized around one chosen weight per row.
The table operationalizes the weighted-sum method by enumerating candidate weights before mapping results into objective space.
Left graph shows axes OBJ_1 and OBJ_2, a purple curve through white points labeled non-dominated Solutions, yellow points labeled dominated solutions, and a red point labeled unattainable Solution.
Weight labels =.99, .9, .75, .5, .25, .1, .01, .001 appear beside the graph, and a table below has headers , J, OBJ_1, OBJ_2.
horizontal axis OBJ_1
vertical axis OBJ_2
purple Pareto-front curve
white non-dominated points
yellow dominated points
red unattainable point
weight labels
table with columns , J, OBJ_1, OBJ_2
boxed formula J=(·OBJ_1)+()·OBJ_2
The instructor points first to the top-left region while discussing minimizing OBJ_1.
He then points toward the bottom-right region while discussing minimizing OBJ_2.
Later he traces the purple curve and points to the red and yellow regions to classify solution types.
The axes remain OBJ_1 horizontally and OBJ_2 vertically.
The purple curve remains the visual boundary connecting the white points.
The boxed weighted-sum formula stays visible on the right.
The diagram organizes the lesson around three classes of points—non-dominated, dominated, and unattainable—and ties them to the weighted-sum formulation.
Full-screen title card reads “The pareto front.”
title text “The pareto front”
The lecture view is replaced by a title card.
No new mathematical notation is introduced during the card itself.
The card marks a transition from the general conflict discussion to the specific concept of the Pareto front.
On the right side of the board the instructor draws a new coordinate system with OBJ_2 vertical and OBJ_1 horizontal, sketches a dashed frontier curve, writes “feasible,” and adds small circles in the feasible region.
An orange overlay at 00:02:17.000-00:02:29.000 reads “we may get first, decent solutions, but we want to improve so they get closer to the pareto front.”
The exact final positions of all added circles are not fully legible frame by frame.
new OBJ_1/OBJ_2 axes
dashed frontier curve
label “feasible”
several small circles in the feasible region
orange explanatory overlay
He draws the axes after saying “let’s go back and look at this graph.”
He sketches the dashed curve as the boundary of attainable solutions.
He writes “feasible” in the region above/right of the curve.
He adds multiple small circles representing initial decent solutions.
He indicates that these solutions should move closer to the frontier.
The axes still represent OBJ_1 and OBJ_2.
The frontier-like dashed curve remains below/left of the feasible cloud.
This visual sequence supports the methodological point that one often starts with feasible but suboptimal solutions and improves them toward the Pareto front.
A white overlay box appears at upper left listing "heuristics:", "simulated annealing", "tabu search", "genetic algorithms", and "...".
White text overlay box
List of heuristic method names
The overlay appears while the speaker is naming heuristic methods.
It disappears after a few seconds.
The underlying blackboard graphs remain visible.
The lecturer continues speaking over the overlay.
The overlay reinforces the spoken list of example heuristic techniques used for multi-objective search.
The lecturer draws several small circles just inside the dashed curve on the right-hand graph.
During this drawing he says the solutions are not on the Pareto front but are near it, and that one wants to populate as many possibilities as possible.
Exact count of newly drawn circles is approximate because some overlap and the hand occludes parts of the plot.
Right-hand objective-space graph
Dashed Pareto-front curve
Multiple small circles near the curve
Lecturer's chalk hand
New circles are added sequentially near the inner side of the dashed frontier.
The density of plotted candidate solutions increases.
The axes labels "obj 1" and "obj 2" remain fixed.
The dashed frontier itself remains the reference boundary.
The region label "feasible" stays in place.
The animation visually represents feasible solutions produced by heuristics that are close to, but not exactly on, the Pareto front.
The lecturer points to individual drawn circles while discussing presenting a solution to a stakeholder.
He says a solution like this one may not be optimal but can still show what the structure of the solution would look like.
Individual small circles near the Pareto front
Lecturer's pointing hand
Attention shifts from the whole cluster to particular example points.
The speaker uses one point as a representative candidate solution.
The overall cluster remains visible near the dashed frontier.
The graph axes and labels do not change.
This visual emphasis supports the claim that a single near-front solution can be communicated as a concrete, though not perfectly optimal, proposal.
A second graph remains visible on the left with labels including "dominated solutions", "non-dominated Solutions", and "unattainable Solution".
The lecturer does not verbally develop the left graph within this clip.
Left graph
Curve with marked points
Labels "dominated solutions", "non-dominated Solutions", "unattainable Solution"
The left graph stays in view throughout the clip.
Its labels remain readable in broad outline.
The left diagram provides contextual terminology for Pareto-related concepts, but the spoken content in this clip focuses mainly on the right-hand objective-space sketch and heuristic outcomes.
The lecturer says the objectives are generally conflicting, but not always.
The board note reads “generally conflictive but not always.”
One might think that having multiple objectives always means the objectives must oppose each other.
The video explicitly states that multiple objectives are generally conflicting, but not always.
The lecturer describes mapping solutions for different weights and obtaining a trade-off curve.
The graph shows multiple solution points along a curve rather than one isolated optimum.
A learner may expect a single best solution even when several objectives are present.
In the presented weighted approach, different weights lead to different solutions, producing a curve of trade-offs instead of one universal optimum.
At 00:00:43.000-00:00:49.000 he says dominated solutions are “not very good.”
One might hear “not very good” as a vague aesthetic judgment.
In the context of the board, dominated solutions are specifically the attainable points that are not on the Pareto front, i.e., they are inferior trade-off points relative to non-dominated ones.
At 00:00:31.000-00:00:49.000 he distinguishes the red unattainable solution from the other points called dominated solutions.
The red point is drawn below/left of the frontier, while yellow dominated points are elsewhere in the plotted space.
One might conflate any “bad” point with being dominated.
The clip separates an unattainable solution, which cannot be found at all, from dominated solutions, which are attainable but not on the Pareto front.
The speaker explicitly counters the idea that only exact Pareto-front membership matters by saying that if heuristics do not get there, having many very close solutions is still what you want.
One might think that if a heuristic does not land exactly on the Pareto front, the resulting solutions are not worth keeping.
The lecturer states that many feasible solutions very close to the Pareto front are still desirable and practically useful.
The speaker says a solution may not be perfectly optimal but can still be presented to a stakeholder to show the structure of the solution.
A solution that is not perfectly optimal might be dismissed as useless for decision making.
According to the lecturer, such a solution can still be valuable for explaining what a good solution structure looks like to a stakeholder.
After defining multiple-objective optimization, the lecturer introduces the weighted objective method for two objectives.
The boxed formula appears beneath the multi-objective heading and objective list.
The weighted-sum formula is presented as a method for handling a multiple-objective problem by collapsing it into a single scalar objective.
The lecturer says the basis is to start with a table of different values and compute J.
The table headers include and J, matching the variables in the boxed formula.
The table implements the weighted-sum formula by enumerating candidate weights and recording the resulting objective values.
The lecturer says that mapping solutions for each weight yields a trade-off curve.
The pointer moves between the lower-left table and the upper-left graph.
Solving for multiple weights and plotting the resulting objective pairs produces the trade-off curve shown in the graph.
The lecturer first notes that objectives are generally conflicting, then later explains the resulting trade-off curve.
The graph visually represents improvement in one objective versus the other along the purple curve.
The existence of a meaningful trade-off curve depends on the fact that improving one objective can come at the expense of the other.
The instructor first describes conflicting objectives (00:00:00.000-00:00:15.000), then introduces the Pareto front as the line linking non-dominated solutions (00:00:23.000-00:00:31.000).
Understanding that the objectives trade off is what motivates the need for the Pareto front as the set of best compromise solutions.
At 00:00:23.000-00:00:31.000 he says the Pareto front links all the non-dominated solutions.
The Pareto front is visually and verbally defined as the curve containing the non-dominated solutions.
At 00:00:31.000-00:00:43.000 he calls one point unattainable; at 00:02:05.000-00:02:13.000 he identifies the rest of the region as feasible.
The feasible space consists of attainable points, while the unattainable solution lies outside that attainable region.
The boxed formula J=(·OBJ_1)+()·OBJ_2 is displayed throughout the clip beside the Pareto discussion.
The clip does not explicitly prove the relation between all choices and all Pareto-optimal points.
The weighted-sum formula is the scalarization tool shown for handling the two-objective problem whose optimal trade-offs are depicted by the Pareto front.
The school example is introduced immediately after the general statement that minimizing one objective harms the other.
The school-system example is used to concretize the abstract notion of conflicting objectives.
The speaker first names heuristics and then explains that they may either reach the Pareto front or produce many nearby solutions.
The named heuristic methods are the search techniques whose outputs are then discussed as either exact Pareto-front points or near-front solution sets.
After defining the near-Pareto outcome, the speaker says one wants to populate as many possibilities as possible.
The practical goal of populating many nearby possibilities is part of the broader strategy of working with near-Pareto solutions.
The lecturer moves from describing near-front solutions to using one such solution in a stakeholder explanation.
Near-Pareto solutions are the objects that can later be presented to stakeholders despite not being exactly optimal.
Opening explanation of multiple objective optimization.
Board heading “Multiple - objective optimization” and “more than one objective function.”
Boxed equation J = ( · OBJ_1) + () · OBJ_2.
Overlay “generic formulation (w=weight)”.
Lecturer says three, four, or five objectives are harder to visualize.
Overlay about taking a bi-objective weighted problem because it is easier to visualize.
Explanation of starting with a table of different values.
Visible table headers , J, OBJ_1, OBJ_2.
Lecturer says mapping solutions for different weights yields a trade-off curve.
Purple curve through multiple points in the OBJ_1-OBJ_2 graph.
“Generally these objectives are conflicting, but not always.”
Board note “generally conflictive but not always.”
At 00:00:23.000-00:00:31.000 the instructor defines the purple line as the Pareto front linking non-dominated solutions.
At 00:00:31.000-00:00:49.000 he distinguishes the red unattainable solution from the yellow dominated solutions.
The boxed formula J=(·OBJ_1)+()·OBJ_2 is visible on the board.
At 00:01:32.000-00:02:00.000 the instructor gives the school cost versus student travel example.
At 00:02:17.000-00:02:29.000 he says we typically get feasible but not optimal solutions and gradually improve them to get close to the Pareto front.
Orange overlay states “we may get first, decent solutions, but we want to improve so they get closer to the pareto front.”
At 00:02:05.000-00:02:13.000 he says “all of this here is the feasible space.”
He writes “feasible” in the region above/right of the dashed curve.
Covered · Title card with no spoken mathematical content; it identifies the topic as multiple objective optimization and pareto front.
Covered · Lecturer introduces multiple-objective optimization and lists typical objectives, noting they are generally conflicting but not always.
Covered · The lecturer restricts attention to two objectives for visualization and gives the school example of conflicting goals.
Covered · The weighted-sum formulation J = ( · OBJ_1) + () · OBJ_2 is introduced and explained.
Covered · Transition slide reading “Optimizing different functions”; no new mathematical statement beyond signaling the next section.
Covered · The lecturer explains the weight table, the computation of J, and how mapping solutions for different weights produces a trade-off curve.
Covered · Opening explanation of conflicting objectives with the weighted-sum formula already visible on the board.
Covered · Title card “The pareto front” marks the topic transition; no new spoken mathematics during the card.
Covered · Instructor classifies points on the first graph into non-dominated, unattainable, and dominated solutions.
Covered · Title card “Getting from an acceptable solution to something closer to the pareto front” introduces the next methodological section.
Covered · He redraws the axes and frontier on a second graph and restates the minimization trade-off.
Covered · School-system example explains how cost and student travel move in opposite directions.
Covered · He marks the feasible region, notes difficulty of reaching the front exactly, and illustrates starting from decent feasible solutions and improving them toward the frontier.
Covered · Opening continuation of the Pareto-front discussion leads directly into the naming of heuristics.
Covered · Audio names simulated annealing, tabu search, and genetic algorithms; matching overlay appears.
Covered · Speaker finishes the point that heuristics can sometimes get to an optimum on the Pareto front.
Covered · Main explanation and drawing of many feasible solutions near, but not on, the Pareto front.
Covered · Speaker explains that a near-front but non-optimal solution can still be shown to a stakeholder.
Covered · Closing remark says this topic will be revisited later when heuristic techniques and other optimization methods are discussed.
Reviewed subject paths
A solution that is not perfectly optimal can still be useful because it can be presented to a stakeholder to illustrate the structure of a good solution. It helps decision makers understand what a reasonable candidate looks like, even if optimality is not guaranteed.
Conditions: The solution is understood to be near-optimal rather than exactly optimal.; This use case is framed as presentation and explanation, not proof of optimality.
The weighted-sum scalarization combines two objective functions into a single aggregate objective by multiplying the first objective by a weight and the second objective by the complementary weight .
Conditions: There are exactly two objectives being combined.; The second weight is defined as .; The objective is to minimize or maximize the combined scalar .
The video distinguishes them by their attainability and position relative to the Pareto front. Unattainable solutions are points in the objective-value plane that cannot be achieved by any feasible decision, lying beyond the attainable frontier.
Conditions: The plot is in objective-value coordinates.; The example concerns attainable solutions of a multi-objective problem.; Both objectives are being minimized in the example.
The purple curve represents the trade-off curve in the objective space. It is formed by mapping the optimal solutions obtained as the weight changes, showing how improving one objective comes at the expense of the other.
Conditions: At least two objectives are plotted against each other.; Solutions are generated for multiple weight values.
The Pareto front is the curve that links all non-dominated solutions in the objective-value plot. Visually, it is the lower-left boundary of the attainable set in this minimization problem, representing the best possible trade-offs between the conflicting objectives.
Conditions: Objectives are plotted in the (OBJ_1, OBJ_2) plane.; The example uses minimization of both objectives.
The weight table serves as the computational scaffold for exploring the family of weighted problems. It lists candidate values for the weight , allowing the user to compute the aggregate objective and record the resulting objective values and for each weight choice.
Conditions: Use a set of candidate weights .; Compute for each weight.; Record the resulting solution’s and values.
The lecturer states that if heuristics do not land exactly on the Pareto front, producing many feasible solutions very close to that front is still a desirable outcome. The practical goal is to populate as many nearby possibilities as possible, which provides valuable candidate solutions even if they are not perfectly optimal.
Conditions: This applies when exact attainment of the Pareto front is not achieved.; The solutions discussed are feasible solutions in the plotted objective space.
The lecturer explicitly names simulated annealing, tabu search, and genetic algorithms as examples of heuristic methods that may be used to search for good solutions in multi-objective optimization.
Conditions: These are presented as example methods, not as a complete taxonomy.; The clip does not define how each heuristic works internally.
Continue with related videos in the same subject.
YouTube10:40
YouTube8:20
YouTube6:05